Single and multiple frequency fiber lase

Single and multiple frequency fiber lasers - CaltechTHESIS
CaltechTHESIS
A Caltech Library Service
About
Browse
Deposit an Item
Instructions for Students
Single and multiple frequency fiber lasers
Citation
Dawson, Jay W.
(1993)
Single and multiple frequency fiber lasers.
Dissertation (Ph.D.), California Institute of Technology.
doi:10.7907/65s0-wf19.
Abstract
NOTE: Text or symbols not renderable in plain ASCII are indicated by [...]. Abstract is included in .pdf document.
Single frequency, low intensity noise, widely tunable lasers operating in the 1.5 [...]m region have potential applications in future wavelength division multiplexed optical communications systems, fiber sensor arrays and high resolution spectroscopic measurements. A single frequency fiber laser having these characteristics will be described in detail. The laser cavity contains an erbium doped fiber gain module, fiber isolators to ensure unidirectional travelling wave operation and two fiber Fabry-Perot filters acting in tandem, which provide broadband tunability (1530 nm -1560 nm) combined with stable single frequency operation. Shot noise limited operation of this laser has been observed at frequencies greater than 300 MHz. At lower frequencies (1-300 MHz) the intensity noise has been characterized in terms of sidemode suppression (> 60 dB of minimum sidemode suppression has been realized). Lower still (10 kHz - 1 MHz) the intensity noise is dominated by the laser's relaxation resonance (30 kHz @ 1 mW output, -105 dBc/Hz). The linewidth of this laser has been measured to be less than 4 kHz using a loss compensated recirculating delayed self-heterodyne interferometer (RDSHI). The RDSHI is an improvement over the standard delayed self-heterodyne interferometer in that the effective delay line can be increased by a factor of 30 over the standard method, increasing the resolution by a corresponding amount. The RDSHI also allows measurement of the short term frequency jitter of a laser. In order to reduce laser frequency jitter, the Pound-Drever technique was employed to lock the laser frequency to an external fiber Fabry-Perot. The same technique also permitted the internal mode selection filter to track the laser frequency, completely eliminating residual mode hopping due to thermal length changes of the laser cavity. Finally, fiber laser configurations that allow multiple frequencies to be simultaneously produced in one laser cavity will be described.
Item Type:
Thesis (Dissertation (Ph.D.))
Degree Grantor:
California Institute of Technology
Division:
Engineering and Applied Science
Major Option:
Applied Physics
Thesis Availability:
Public (worldwide access)
Research Advisor(s):
Vahala, Kerry J.
Thesis Committee:
Unknown, Unknown
Defense Date:
11 May 1993
Record Number:
CaltechETD:etd-08272007-084804
Persistent URL:
DOI:
10.7907/65s0-wf19
Default Usage Policy:
No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:
3238
Collection:
CaltechTHESIS
Deposited By:
Imported from ETD-db
Deposited On:
28 Aug 2007
Last Modified:
19 Apr 2021 22:35
Thesis Files
Preview
PDF (Dawson_jw_1993.pdf)
- Final Version
See Usage Policy.
5MB
Repository Staff Only:
item control page
CaltechTHESIS is powered by
EPrints 3.3
which is developed by the
School of Electronics and Computer Science
at the University of Southampton.
More information and software credits
Single and Multiple Frequency Fiber Lasers

Thesis by

Jay W. Dawson

In Partial Fulfillment of the Requirements
for the Degree of
Doctor of Philosophy

California Institute of Technology
Pasadena, California
1993
(Submitted May 11, 1993)

Jay W. Dawson

To
Catherine
and to
my parents

Acknowledgments

It is a great pleasure to acknowledge the support and encouragement of my advisor,
Professor Kerry J. Vahala. He and his group provided the atmosphere and resources
necessary for exciting research and for my growth as an independent scientist.

I would especially like to thank Namkyoo Park, who collaborated with me on much
of the work contained in this thesis. I would also like to thank Steve Sanders who
collaborated with me on the intensity noise measurements contained in chapter three of the
thesis. I have also enjoyed working with Jianhui Zhou, whose experiments in four-wave
mixing have helped to provide a test of the fiber laser's usefulness in high resolution
spectroscopy.

I would also like to thank the other members of Dr. Vahala's group who have
provided support and encouragement during my time at Caltech. These include Dr. John
Lebens, Dr. Michael Newkirk, Dr. Michael Hoenk, Dr. Pete Sercel, Dr. Winston
Saunders, Charles Tsai, Robert Lee and Dave Geraghty. The help of Rosalie Rowe is truly
appreciated.

I am very grateful to the Northrop Corporation whose funding has supported much
of this work. I am especially grateful to Dr. Wes Masenten at Northrop, who has been
very encouraging of this research. In addition, I would also like to thank Calvin Miller at
Micron Optics for use of the narrowband fiber Fabry-Perot used early on in this research. I
would especially like to thank the Ortel Corporation for the loan of the high speed
photodiode used in the linewidth measurements in chapter four.

The encouragement of my parents and the rest of my family will always be
appreciated.

I would especially like to thank my wife Catherine for her love and support.

Sa a

eee

Abstract

Single frequency, low intensity noise, widely tunable lasers operating in the 1.5 fim region
have potential applications in future wavelength division multiplexed optical
communications systems, fiber sensor arrays and high resolution spectroscopic
measurements. A single frequency fiber laser having these characteristics will be described
in detail. The laser cavity contains an erbium doped fiber gain module, fiber isolators to
ensure unidirectional travelling wave operation and two fiber Fabry-Perot filters acting in
tandem, which provide broadband tunability (1530 nm -1560 nm) combined with stable
single frequency operation. Shot noise limited operation of this laser has been observed at
frequencies greater than 300 MHz. At lower frequencies (1-300 MHz) the intensity noise
has been characterized in terms of sidemode suppression (> 60 dB of minimum sidemode
suppression has been realized). Lower still (10 kHz - 1 MHz) the intensity noise is
dominated by the laser's relaxation resonance (30 kHz @ 1 mW output, -105 dBc/Hz).
The linewidth of this laser has been measured to be less than 4 kHz using a loss
compensated recirculating delayed self-heterodyne interferometer (RDSHI). The RDSHI is
an improvement over the standard delayed self-heterodyne interferometer in that the
effective delay line can be increased by a factor of 30 over the standard method, increasing
the resolution by a corresponding amount. The RDSHI also allows measurement of the
short term frequency jitter of a laser. In order to reduce laser frequency jitter, the Pound-
Drever technique was employed to lock the laser frequency to an external fiber Fabry-
Perot. The same technique also permitted the internal mode selection filter to track the laser
frequency, completely eliminating residual mode hopping due to thermal length changes of
the laser cavity. Finally, fiber laser configurations that allow multiple frequencies to be

simultaneously produced in one laser cavity will be described.

Contents
1 Introduction 1
1.1 Optical Amplifiers for Fiber Optic SystemS................cceeeeee 2
1.2 Fiber Lasers... ccc cece cee sete eens neeeeseneneeeassseeeeesenes 6
1.3. Applications of Fiber Lasers............c eee ccs eeeseeeeeee ones 12
1.4 Thesis Outline... cece ceceececeececeeaececeeeseeeceeeenenes 14
2 A Single-Frequency Fiber Laser 25
2.1 | Design Considerations for Optimal Single-Frequency Operation of an
Erbium-Doped Fiber Laser....... ec cceeccceecccseeceeeeeeeeaeeeeeees 25
2.2 Construction of an All-Fiber, Widely-Tunable, Single-Frequency
Erbium-Doped Ring Laser.........cceeccceccceeseeecneeceeeseceeeeeees 37
2.3. Lasing Characteristics of an All-Fiber, Widely-Tunable, Single-
Frequency Erbium-Doped Ring Laset.........cccecceeesecceeeeees 42
3 Intensity Noise Characteristics of a Single-Frequency 54

Fiber Laser

3.1
3.2

Low and Intermediate Frequency Intensity Noise Spectra............ 55
Measurement of the High Frequency Intensity Noise Spectra
Using the Balanced Homodyne TeCHNique..........ceceeessceeeeeeeees 67

vil

Linewidth and Frequency Jitter Measurements Using an 82
Improved Delayed Self-Heterodyne Interferometer
4.1 An Improved Delayed Self-Heterodyne Interferometer for

Linewidth Measurements..............ccccccecsseescceeeeeeeesceseeeneees 83
4.2 Linewidth and Frequency Jitter Measurement of a Single-

Frequency Fiber Laser.............ecceeececcceccnecnseceeeceeneseeeeeeees 96

Frequency Stabilization of a Single-Frequency Fiber Laser 103
5.1 Frequency Stabilization... cee ccccceececeeceeceteeeseeeeees 104
5.2. Application of the Pound-Drever Technique to a Single-

Frequency Fiber Laser..........ccc cece esse cteecceeceeeceeeseeeees 108
Multiple Wavelength Fiber Lasers 120
6.1 Spectral Hole Burning in Erbium Doped Fiber Amplifiers........... 121
6.2 Co-Lasing in an Electronically Tunable Fiber Laser.................. 128
6.3 N-Frequency Lasers... cece cec cece ecnececceseeeeeneeteneaeeneaes 135 '

Vili

List of Figures

1.1
1.2
1.3
1.4

2.1
2.2
2.3
2.4
2.5
2.6

2.7

2.8
2.9
2.10
2.11

Schematic of a typical erbium doped fiber amplifier.................. 4

Passively mode-locked fiber laser..............c..eeceeeeeeeeeeeeeeeeees 8

Examples of single-frequency fiber laser designs.................0.+. 10
Four-wave mixing in a semiconductor optical amplifier.............. 13
Gain characteristics of a commercial EDFA... eee eee 27
Experimental set-up for measuring the gain in an EDFA.............. 28
Measured gain characteristics of an EDFA... eeeceeeeeee 30
Tandem fiber Fabry-Perot filter concept... eee ceeeeeeeeees 36
Erbium-doped fiber ring laser... ee eeee eens Veseeeeees 38

Output spectrum of the fiber laser as seen with a Newport
Research Super Cavity.........cccceccecssececssececeeceeeneeeeseesaeenees 43

Output spectrum of the fiber laser as seen with a Newport

Research Super Cavity (close-up)............ccccccccccssesseeeeeeeenees 44
Wavelength as a function of voltage applied to the BB FFP......... 46
Output POWEr VS. PUMP POWER.............. ceceessessseeeeeeeeeeneees 47
Output power vs. wavelength........... 0. cceeececcceceseteessseeeees 48

Sidemode suppression vs. wavelength... cceecccssseseeees 49

3.1

3.2

3.3

3.4

3.5

3.6

3.7

3.8
3.9

3.10

3.11

3.12

4.1

4.2

4.3

4.4

Experimental set-up for measuring the low frequency intensity
NOISE SPECITA..... ce cececeecscecees ee ecesteceeeeeeesceeeeeseeneeeeeenaeerees
Relative intensity noise spectrum of the fiber laser from 10 kHz

tO 1 MHZ... cecceeteceec tec eceeereceeeneeeseeeeeseseesteeeesenees
Theoretical fit of figure 3.2 data to equation 3.2..........ccesee
Relaxation oscillation frequency vs. fiber laser output power.......
Intensity noise spectrum of the fiber laser from 1 to 20 MHz.......
Intensity noise spectrum of the fiber laser from 1 to 400 MHiz......
Balanced homodyne system to determine laser noise power
relative to the SQL (shot noise flOOP)................-ccesceeeeeeeseeeees
Noise power (arbitrary units) vs. laser POWET.............cccseeees
Noise power relative to the SQL as a function of spectrum
analyzer frEQUENCY......... cece ce cee ce cee eee ececeeeeeeeeeeecneeeesenees
Noise power relative to the SQL as a function of spectrum
analyzer frequency (50% output coupling)............. pehesssesereeeees
Noise power relative to the SQL as a function of spectrum
analyzer frequency (90% output coupling)... eens
Noise power relative to the SQL as a function of spectrum

analyzer frequency (10% output coupling)... eeeeeeeees

The conventional delayed self-heterodyne interferometer.............
The loss-compensated recirculating delayed self-heterodyne
INtETFETOMECLET......... 0 cece ee eee eee e eee e eee e cence eeceeeeeeseseeeeeeeenes
Photocurrent power spectra of the recirculator in the case of no
EDFA in the recirculator................. seseeeseaneaseeeseseeseneestneeees
Photocurrent power spectra of the recirculator in the case of an

EDFA in the recirculator............c.ccccccccssescescsesccscescscnscseeees

61
63

69
71

4.5

4.6

4.7

4.8

4.9

4.10

5.1

5.2

5.3

5.4

5.5

5.6

6.1

6.2

6.3

6.4

Measured linewidth of the fiber laser as a function of order.......... 92
RDSHI output without the delay line... eee ee eeeees 94
RDSHI output with the EDFA and without the EDFA................ 95
Typical spectrum of the RDSHI at the 10th order... 97

Measured linewidth of the fiber laser between the wavelengths
Of 1531 and 1538 MMe... cece eeeee sees sete eeeeenseeeeees 98

One possible case of laser frequency VS. tiME.............cccceeeee 100

Pound-Drever technique for frequency locking to an external
Fabry-Perot Cavity.......c.cccecccceeeeeeeeecseceseeteseesssensaeeeeeueees 106
An internal tracking filter is required to achieve active locking
to an external reference... ce cece cee eenceeeecenaneeeeeeees 109

Experimental set-up for active stabilization of the fiber laser

FLEQUENCY.... eee cece eee cece nee e cee eee ceeceeeeaeesetensneeseeeaee 110
Error signal from the internal NB FFP... cesssseeeseeeeee 113
Error signal from the external reference NB FFP..........00...00. 115

Experimental set-up to stabilize two fiber lasers to the same

reference FRP io... ccc ccecc cece cee ceeseccecccessecueeceesccenensens 116

Energy level diagram showing the Stark components of the upper and
lower energy levels of the EDFA... ..cccccceeeesescceseeeeeees 122
Experimental set-up for measuring gain cross-saturation and

spectral hole burning... cece cee seeeceeeeeteeeeecensenees 124
EDFA small signal gain vs. wavelength in the case of no other

signal; signal at 1536.6 nm; signal at 1537 nm... eee 126
EDFA small signal gain vs. wavelength for varying signal

6.4

6.5

6.6

6.7
6.8
6.9
6.10
6.11
6.12

6.13

xi
signal; signal at 1536.6 nm; signal at 1537 nm.................cc08

EDFA small signal gain vs. wavelength for varying signal

Dual -frequency fiber laser with one gain medium and two
independent gain Mei... cccsessceccecaeeseeseceeeeeeeeseees
Typical output spectrum of the Newport Research Super Cavity
spectrum analyzer, showing two co-lasing modes separated by
several nanometers in wavelength but folded over by the 6 GHz
FSR of the Super Cavity.......ccccccccccccccccsccssesesscsssessssenees
Experimental tuning data from the configuration in figure 6.5a.....
Experimental tuning data from the configuration in figure 6.5b......
Theoretical calculations of Mach-Zehnder throughput.................
Eight-channel laser configuration based on a linear cavity............
Eight-channel laser configuration based on a ring cavity.............
Spectrum of the multiplexed output port of a six-channel

TING ASOT... ieee eee eeceeceecusceeceseeeeeceseececesssseuseeseeeeces

N-frequency fiber laser (proposed)..........ccccccccccccscsssssseeceees

Chapter 1

Introduction

The purpose of this thesis is to describe the development and characterization of
single and multiple frequency fiber lasers. Single frequency fiber lasers have several
attractive properties which will be useful in a high resolution spectroscopy system or a
wavelength division multiplexed communications system. These include narrow linewidth,
broadband tunability and inherent compatibility with optical fiber. Fiber lasers are
inherently compatible with optical fiber in two ways. First, the light is generated in the
fiber making coupling into an optical system a simple matter of a fusion splice. Second,
the lasers are based on fiber amplifiers and are by default compatible with the wavelength
range at which those amplifiers operate.

This introductory chapter is meant to provide perspective for the discussion of the
single-frequency fiber laser. Fiber amplifiers, upon which the fiber laser is based, will be
discussed first. Other work in the field of fiber lasers and some applications in which fiber
lasers have already made a contribution will then be discussed. Finally, a brief outline of

the thesis will be provided.

Sree

1.1 Optical Amplifiers for Fiber Optic Systems

There are two types of optical amplifiers commonly considered for use in fiber optic
Systems: semiconductor based optical amplifiers and rare-earth ion doped optical fiber
amplifiers. (Other types of amplifiers such as Raman amplifiers have been demonstrated
[1]; however, the broader bandwidths and simpler pumping schemes of rare-earth ion
doped fiber amplifiers have virtually eliminated interest in these alternatives.) Although
both of these types of amplifiers may now be purchased commercially, they are still the
subject of active research.

Semiconductor optical amplifiers are currently being studied as a means of
compensating the high loss inherent in integrated photonic devices [2]. Semiconductor
amplifiers have a number of impressive qualities including a large gain bandwidth [3] and
high saturation output power [4]. However, the fast recovery time of the semiconductor
gain media leads to large cross-talk if two or more signals are amplified simultaneously [5].
This makes them unsuitable for use as "in-line" amplifiers in future optical communications
systems. Future systems will no doubt utilize multiple wavelengths. There is, however,
one "silver lining" to the crosstalk problem and that is that it may be possible to use the
cross talk effect to construct an optical frequency shifter [6]. Such a device would be very
useful in future wavelength division multiplexed communications systems [7].

Rare-earth ion doped optical fiber amplifiers have, in a very short amount of time,
revolutionized the way the telecommunications industry thinks about fiber optic systems.
A rare-earth ion doped fiber was first conceived by Snitzer [8] in 1961. However, major
interest in fiber amplifiers did not develop until as late as 1985 when Poole et al. [9]
demonstrated rare-earth ion doping of single mode optical fibers. Demonstration of a
neodymium fiber laser end pumped by a laser diode was published simultaneously [10].

Progress since then has been very rapid in theoretical modelling [11-16], diode laser pump

sources [17,18,19], investigation of dopants [20-24] and in actual amplifier demonstrations
[25, 26, 27]. Today, it is rare to see a proposed fiber optic system that does not include a
rare-earth doped fiber amplifier as an essential component.

This thesis focuses on fiber lasers based on erbium doped fiber amplifiers (EDFA).
EDFAs are currently the most advanced in terms of development of the rare-earth ion doped
fiber amplifiers. Several companies now make EDFAs as commercial products. An EDFA
for telecommunications applications (Figure 1.1) consists of a length of single mode
erbium doped fiber, two wavelength division multiplexers and a laser diode for pumping.
A typical fiber core diameter for the erbium doped fiber is 4 Um, with an erbium
concentration of 50 ppm. The core is codoped with an index-raising codopant, which
forms the single mode waveguide. This codopant has been found to affect the spectral
properties of the amplifier gain [28, 29, 30]. Aluminium in combination with germanium
has been found to provide the broadest amplifier gain bandwidth and is now most
commonly used.

The wavelength division multiplexer (WDM) is a fused fiber coupler that acts as a
dichromatic mirror. The WDM combines pump light from a laser diode with signal light to
be amplified. The erbium doped fiber is spliced to the WDMs using a thermally expanded
core fusion splicing technique [31] to reduce losses due to mismatch between the mode
field diameter of telecommunications fiber and the erbium fiber. The second WDM may be
employed simply to couple pump light out of the system or to allow bidirectional pumping
of the amplifier. If the amplifier is pumped from only one direction, counter-directional
pumping is usually preferred because slightly higher gain is attainable [12].

A 980 nm or 1480 nm laser diode is normally used as a pump source. 980 nm
pumping is preferred due to a lower amplifier noise figure than that obtained with a 1480

nm pump source [32]. Other pump wavelengths are also possible including, 820 nm [33],

Input
Signal
LD
WDM
ras rs
WDM

Amplified
Signal

teectaseneae
sadam

Figure 1.1: Schematic of a typical erbium doped fiber
amplifier. WDM: wavelength division multiplexer, LD:
pump laser diode, X: thermally diffused core fusion
splice, Er>* erbium doped fiber

514 - 532 nm [34] and 665 nm [35]. Additionally, for erbium doped fiber codoped with
ytterbium pumping throughout a broad band from 800 nm to 1100 nm is possible [36].
The ytterbium acts as a broadband absorber and efficiently transfers light into the upper
level of the lasing transition.

A typical amplifier consists of a 980 nm pump diode with a maximum output of 30
mW and 25 m of aluminium and germanium codoped erbium doped fiber (50 ppm erbium).
Such an amplifier can provide up to 40 dB of small signal gain at the gain peak (1532 nm
for aluminum and germanium codoped fiber, 1537 nm for germanium only codoping) and
provide a saturation output power of 10 mW. The bandwidth can exceed 50 nm (1520 nm
- 1570 nm) [30]. EDFAs made with ytterbium codoped fiber and pumped with a
neodymium YAG laser at 1060 nm have been demonstrated with saturation output powers
of nearly 1 W [37].

The erbium gain medium is interesting in that the broad gain bandwidth comes from
a combination of Stark splitting of the upper and lower laser transition levels and
occupation of multiple sites in the glass by the triply ionized erbium atoms. From this, one
might expect the amplifier's gain to saturate inhomogeneously; however, it saturates
homogeneously [38]. Homogeneous saturation is due to rapid cross relaxation between the
closely spaced Stark levels [39]. This cross relaxation occurs at a finite rate leading to very
weak (about 1 dB depth, 1 nm width) spectral gain hole burning allowing simultaneous
lasing of multiple frequencies (40, 41]. However, for most situations, a simple
homogeneous model of the gain is a sufficient description.

For the triply ionized erbium atom in free space, the lasing transition that we are
interested in 4113/2 --> 4115/2 is electric dipole forbidden. However, the ligand field in the
glass fiber host mixes the free atom energy states to allow a weak electric dipole transition
between these states. The weak electric dipole nature of this transition leads a long upper

level lifetime, 12 ms [42]. Thus the EDFA gain responds very slowly in comparison to the

——

“ewer

———

proposed data rate of fiber optic systems, leading to low crosstalk between multiple signals
[43].

Current areas of research and development in EDFAs include polarization
preserving erbium doped fiber [44], high concentration erbium doped fiber [45] and 980

nm laser sources with output powers greater than one watt in a single transverse mode

[18].

1.2 Fiber Lasers

As mentioned previously, the first fiber laser can be attributed to Snitzer [8].
However, current widespread interest in fiber lasers began with Mears, et al., [10] when
fiber amplifiers were developed. Presently there are several areas of active research in fiber
lasers: mode-locked fiber lasers, upconversion fiber lasers and single frequency fiber
lasers.

Both active and passively mode-locked fiber lasers have been demonstrated.
Actively mode locked fiber lasers usually employ a ring configuration and an electro-optic
modulator [46, 47, 48]. They generate picosecond pulses with repetition rates of up to 30
GHz [49]. Mode locked fiber lasers with cavities constructed from all polarization
maintaining fibers have been demonstrated [50]. Stabilized systems that actively control
either the drive frequency to the electro-optic modulator or the cavity length, to keep the
cavity mode spacing resonant with the modulation, have been demonstrated [51]. In
addition a multiple wavelength mode-locked fiber laser has recently been demonstrated
[52].

There is great interest in passively mode-locked erbium doped fiber lasers. The
most popular scheme for obtaining passive mode-locking of a fiber laser is through the use
of a nonlinear amplifying loop mirror (NALM) [53-56]. By splicing together the two

output ends of a 3 dB fused fiber coupler, a loop mirror or Sagnac interferometer may be

constructed. All of the light injected into one port of this loop mirror is reflected back out
the input port due to the phase shift inherent in the fused fiber coupler, the identical path
lengths traversed by the counterpropagating beams and constructive interference.
However, if a gain or loss element is placed at an asymmetric location in the loop, the light
propagating in one direction will have more power than the light propagating in the
opposite direction. For a large enough difference in power, self phase modulation in the
fiber will result in light travelling one direction to be phase-shifted with respect to light
travelling the opposite direction. For a phase difference of x the reflector becomes a
transmitter. Thus the NALM acts as a transmitter for high intensity signals and a reflector
for low intensity signals.

The "figure eight" laser or passively mode-locked fiber laser is based on this
principle (Figure 1.2). A loop mirror is included in a unidirectional ring. Only high
intensity light can propagate completely around the ring. The laser is forced to mode-lock
in order to achieve the necessary intensity to oscillate. Both square pulses and solitons
have been observed [57] using this configuration. Pulse widths on the order of 300
femtoseconds have been obtained [57]. However, control of the pulse repetition rate is
somewhat difficult [58]. In order to get high repetition rates, mode-locking must occur at a
high harmonic of the cavity's fundamental repetition rate [55]. More research is needed in
this area before the figure eight laser will be a practical device.

Upconversion is a process where an atom absorbs two or more photons then emits
a photon of higher energy than any one of the initial photons. Rare earth ions make
excellent upconvertors because the large number of closely spaced energy levels translate
into an even larger number of available transitions in the infrared. Thus by multiple
absorption of infrared photons, an electron can ascend to a high energy level and then fall

almost all the way back to the ground state in a single jump emitting a photon of larger

Figure 1.2: Passively mode-locked fiber laser, ISO: isolator,
WDM: wavelength division multiplexer, LD: laser diode,
50/50: fused fiber coupler, OC: output coupler

energy than any one of those absorbed. Based on this principle, it is possible to construct
lasers that are pumped by efficient semiconductor lasers, but lase at visible wavelengths. A
recent review article discusses these processes in more detail [59]. Rare earth ion doped
fibers are excellent candidates for upconversion lasers, because the pump light intensity can
be kept quite high over a long interaction length. Several such upconversion fiber lasers
have been demonstrated [60, 61, 62]. Erbium has been shown to lase in the green 546 nm,
when placed in a fluorozirconate fiber and pumped at 801 nm [62]. A praseodymium
doped fluoride fiber sensitized with ytterbium was shown to lase in the red at 635 nm when
pumped in the 810-860 nm region [60]. Another praseodymium doped flouride fiber was
simultaneously pumped at 1010 nm and 835 nm and exhibited lasing in the red (635 nm,
605 nm), green (520 nm) and blue (491 nm) [63]. Holmium and thulium doped fiber
lasers that are pumped in the red or infrared and lase in the visible region have also been
demonstrated [64, 65]. These upconversion fiber lasers have potential to be compact
sources of visible coherent light for applications such as displays and data storage.

Single frequency fiber lasers have been demonstrated based on neodymium and
erbium doped fibers [66-76]. Many different cavity configurations have been tested
(Figure 1.3). These include ring geometries with some kind of broad band filter such as a
fiber Fabry-Perot [74], an acousto-optic tunable filter [69], a liquid crystal filter [68] and a
bulk optic filter [71, 75]. So far the laser geometry discussed in this thesis, utilizing two
fiber Fabry-Perot filters acting in tandem, has been the only ring geometry fiber laser to
achieve stable single frequency operation without rapid mode hopping over a range of
several GHz.

Standing wave cavity designs must deal with the effects of spatial hole burning,
when attempting to achieve single frequency operation. Three means have been
demonstrated to deal with this effect. One design uses a bulk optic grating as a mirror [75].

The very narrow feedback properties of the grating combined with a short cavity allow

(a) oc
ISO FIL
(b) oc
3+
R | Er ~
i |
(c) Er 3+

Figure 1.3: Examples of single frequency fiber laser designs,
(a) ring geometry, (b) linear geometry, (c) bulk geometry,
ISO: isolator, OC: output coupler, Er3* gain media, FIL:
filter, R: reflector, GR: grating reflector,

eee Sea

selection of a single longitudinal mode of the cavity. If no other modes have feedback,
spatial hole burning will not be an impediment to single frequency operation. This system
allows broad band tuning of the laser combined with stable single frequency operation.
However, due to the large losses inherent with this system the lasing threshold is high.
Also, its optical alignment must be monitored if the laser is to operate for long periods of
time. An all fiber laser does not suffer from this long term alignment problem.

The second method uses a very short (approximately 2 cm) piece of erbium doped
fiber combined with intracore Bragg reflectors written with ultraviolet light [66, 76]. This
method has several advantages. The system is all fiber, lases stably in a single longitudinal
mode, is compact and potentially inexpensive to manufacture. However, the tuning range
is limited. These "Bragg reflector" fiber lasers will probably find some applications in
sensors [77].

The third method uses a phase modulator at the end of the cavity to sweep the
spatial standing wave pattern relative to the fixed gain ions [67]. If the field is swept rapidly
relative to the stimulated lifetime of the gain media, the spatial pattern in the gain can be
eliminated, eliminating spatial hole burning. However, dithering the cavity length also
dithers the laser frequency resulting in a linewidth on the order of the free spectral range of
the cavity.

Narrow linewidth is one of the main attractions of the ring cavity configuration.
Because unidirectional ring lasers are travelling wave devices, spatial hole burning is
eliminated. This allows the construction of longer cavities. The natural or Schawlow-
Townes linewidth of a laser is inversely proportional to the square of the resonator length.
It is possible to construct single frequency fiber lasers with cavity lengths on the order of
50 m [74]. At the present time, the linewidth of the laser is dominated by fluctuations in
the cavity length due to other effects [78]. Frequency stabilization schemes can be
employed to eliminate mode hopping and the thermal drift of the cavity length which cause

drifts in the lasing frequency [79].

1.3 Applications of Fiber Lasers

It is hoped that fiber lasers may play a role in several areas. These include use of
single frequency and mode-locked fiber lasers in fiber optic communications systems; high
resolution frequency and time domain spectroscopy; and fiber sensors. Already, single
frequency fiber lasers made with the intra-core Bragg grating technique have been
demonstrated as temperature and strain sensors [77]. A single frequency fiber laser based
on a bulk grating and frequency stabilized to acetylene has been used to as a tool for high
resolution spectroscopy in the study of rubidium [80].

Three single-frequency fiber lasers have been employed in an experiment in which
two lasers of different frequencies (f1, f2 (f2>f1)) are sent through a multiple quantum well
semiconductor travelling wave optical amplifier [81] (figure 1.4). A nonlinear component
in the amplifier's gain arising from the effects of interband and intraband dynamics mixes
the two optical fields generating new frequencies at fl - (f2-f1) and f2 + (f2 - f1). The
power in these new frequencies is small, so a third fiber laser is employed to allow
heterodyne detection of the new frequency.

The advantages of the erbium doped fiber lasers in this spectroscopic application are
wide tunability in combination with narrow (< 2 kHz, on time scales < 1 ms) linewidth.
The wide tunability allows observation of new frequencies created up to 1 THz away from
the pump frequencies. The narrow linewidth of the lasers allow one to narrow the
resolution bandwidth of the detection system down to the 10 kHz range, thus lowering the
noise floor of the measurement system in comparison to other systems which might employ

broader linewidth semiconductor lasers [82].

seeceeees

Fiber Lasers Fiber Laser
(Pump,Probe) (Local Osc.)

50/50

= _——
ISO ISO

AMP

SPEC

Figure 1.4: Four-wave mixing in a semiconductor optical
amplifier, 50/50 fused fiber coupler, ISO: isolator, TWA:
travelling wave amplifier, SPEC: spectrometer, AMP:
amplifier, PD: photodiode, SA: spectrum analyzer,

1.4 Thesis Outline

This thesis describes the development and characterization of a single-frequency,
widely tunable erbium doped fiber ring laser. Chapter 2 discusses in detail the design
considerations and basic principles of operation of the single frequency fiber laser. Basics,
including the gain media and the selection of the ring configuration, are considered. The
tandem fiber Fabry-Perot filters, their interaction with each other and their effect on side
mode suppression and tuning are discussed. Experimental verification of single frequency
operation, tuning range, threshold pump power, efficiency and output power are described.

Chapter 3 looks at the intensity noise of the laser. The intensity noise is measured
in three spectral regimes, noise at frequencies less than the free spectral range of the laser
cavity, noise primarily due to nearby ring cavity sidemodes beating with the main lasing
mode and noise at frequencies large compared with the narrowband fiber Fabry-Perot filter
bandwidth. It is shown that the spectral content of the intensity noise at frequencies greater
than twice the narrowband fiber Fabry-Perot filter bandwidth can be reduced to the shot
noise limit by proper choice of output coupling and placement of the narrowband fiber
Fabry-Perot filter between the gain media and the output coupler. This result is verified
experimentally using a balanced homodyne measurement system.

Chapter 4 looks at the fiber laser's phase and frequency noise. Because of the
narrow linewidth of the fiber laser, it was not possible to use a standard delayed self-
heterodyne interferometer to measure the laser's linewidth as the amount of fiber required
for the delay line would be prohibitive (200 km for 1 kHz resolution). A new technique
was developed that uses a loss compensated recirculating delay line to measure the laser
linewidth to 1 kHz resolution using only 11 km of fiber and to allow measurement of short

term (< 2 ms) frequency jitter of the fiber laser. This new technique will be described in

detail. Experimental measurements of the laser's phase and frequency noise using this
technique will then be presented.

Chapter 5 discusses frequency stabilization of the fiber laser. The linewidth and
frequency jitter measurements described in chapter 4 indicated that the frequency jitter time
scale is slow enough so that a frequency control loop of relatively low bandwidth (< 20
kHz) would be sufficient to stabilize the laser frequency. The Pound-Drever technique was
used to lock the laser frequency to the transmission peak of an external fiber Fabry-Perot
cavity. The Pound-Drever technique for frequency locking of a laser to an external cavity
will be reviewed. Implementing frequency stabilization to an external reference required
locking the internal mode selector to the external reference as well as locking the cavity
mode to the external reference. Changes in design necessary to achieve this result will be
discussed. Locking of two lasers to the same external cavity will then be considered.
Relative frequency stability of two lasers locked to the same cavity will be investigated
using this technique and compared to the measurements from the loss compensated
recirculating delayed self-heterodyne interferometer described in chapter 4.

The overall cost of this laser system is quite high (currently around $30,000).
However, one of the major cost factors, that of cost per mW of 980 nm pump sources is
rapidly falling [19]. Erbium fiber is also becoming more readily available at lower prices.
Soon the majority of the cost of a laser such as this one will be contained in the mode
selection and frequency stabilization components. Multiple frequency fiber lasers offer the
possibility of effectively distributing the cost of these components among several fiber
lasers, thus lowering the cost per laser.

Chapter 6 looks at the possibility of multiple frequency fiber lasers. It is possible,
using wavelength division multiplexing technology, to generate multiple single longitudinal
modes in the same cavity at fixed frequency spacings. There are two ways to achieve this:
by giving each wavelength its own gain section or by carefully matching the loss at each

wavelength to the gain at that wavelength and employing a single gain module (very weak

spectral hole burning allows simultaneous lasing of more than one wavelength). Two
different experiments were conducted that investigated multi-frequency lasing. One
involved a tunable dual-frequency laser with one or two gain media. The other involved
investigation of wavelength division multiplexing technology and the construction of a six-
frequency single gain module device. The weak spectral hole burning effect is investigated
using an erbium doped amplifier and two tunable fiber lasers. Finally, an N-frequency

fiber laser is proposed based on this work and the work in the preceding chapters.

References

[1] R. H. Stolen, E. P. Ippen, Appl. Phys. Lett. 22, 276 (1973)

[2] T. L. Koch, U. Koren, JEEE J. Quantum Electron. QE-27, 641 (1991)

[3] M. Bagley, G. Sherlock, D. M. Cooper, L. D. Westbrook, D. J. Elton, H. J. Wickes,
P. C. Spurdens, W. J. Devlin, Electron. Lett. 26, 512 (1990)

[4] G. Eisenstein, U. Koren, G. Raybon, T. L. Koch, J. M. Wiesenfeld, M. Wegener, R.
S. Tucker, B. I. Miller, Appl. Phys. Lett. 56, 1201, (1990)

[5] G. Eisenstein, R. S. Tucker, J. M. Wiesenfeld, P. B. Hansen, G. Raybon, B. C.
Johnson, T. J. Bridges, F. G. Storz, C. A. Burrus, Appl. Phys. Lett. 54, 454 (1989)

[6] K. Inoue, Electron. Lett. 23, 911, (1987)

[7] A. A. M. Saleh, paper ThC1 in Optical Fiber Communication 1992 Technical Digest

Series Volume 5 (Optical Society of America), 199 (1992)

[8] E. Snitzer, Phys. Rev. Lett. 7, 444 (1961)

[9] S. B. Poole, D. N. Payne, M. E. Fermann, Electron. Lett. 21, 737 (1985)

[10] R. J. Mears, L. Reekie, S. B. Poole, D. N. Payne, Electron. Lett. 21, 738 (1985)
[11] J. R. Armitage, Applied Optics 27, 4831 (1988)

[12] P. R. Morkel, R. I. Laming, Optics Letters 14, 1062 (1989)

[13] E. Desurvire, J. R. Simpson, J. Lightwave Tech. 7, 835 (1989)

14 E. Desurvire, J. Ligthwave Tech. 8, 1517 (1990)

[15] J. R. Armitage, JEEE J. Quantum Elec. QE-26, 423 (1990)

[16] E. Desurevire, EEE Photon. Tech. Lett. 2, 208 (1990)

[17] S. Uehara, M. Okayasu, T. Takeshita, O. Korgure, M. Yamada, M. Shimizu, M.

Horiguchi, Optoelctron. Devices Technology 5,71, (1990)
[18] D. Mehuys, D. F. Welch, L. Goldberg, Electron. Lett. 28, 1944 (1992)

[19] T. Moss, Photonics Spectra 26, 28 (1992)

[20] R. J. Mears, L. Reekie, I. M. Jauncey, D. N. Payne, Proceedings OFC Reno,

Nevada (1987)

[21] E. Desurvire, J. R. Simpson, P. C. Becker, Optics Letters 12, 888 (1987)

[22] I. Sankawa, H. Izumita, S. I. Furukawa, K. Ishihara, JEEE Photon. Tech. Lett. 2,
422 (1990)

[23] Y. Dusteste, M. Monerie, J. Y. Allain, H. Poignant, Electron. Lett. 27, 626 (1991)

[24] S. F. Carter, D. Szebesta, S. T. Davey, R. Wyatt, Electron. Lett. 27, 628 (1991)

[25] R. C. Steele, G. R. Walker, IEEE Photon. Tech. Lett. 2, 753 (1990)

[26] D. A. Fishman, J. A. Nagel, T. W. Cline, R. E. Tench, T. C. Pleiss, et al., JEEE
Photon. Tech. Lett. 2, 753 (1990)

[27] G. R. Walker, R. C. Steele, N. G. Walker, J. Ligthwave Tech. 8, 1409 (1990)

[28] M. Nakazawa, Y. Kimura, Electron. Lett. 27, 1065 (1991)

[29] M. Ohashi, K. Shiaki, Electron. Lett. 27, 2143 (1991)

[30] C. G. Atkins, J. F. Massicott, J. R. Armitage, R. Wyatt, B. J. Ainslie, S. P. Craig-
Ryan, Electron. Lett. 25, 910 (1989)

[31] C. P. Botha, Electron. Lett. 24, 243 (1988)

[32] M. Yamada, M. Shimizu, M. Okayasu, T. Takeshita, M. Horiguchi, Y. Tachikawa,
E, Sugith, JEEE Photon. Tech. Lett. 2, 205 (1990)

[33] M. Nakazawa, Y. Kimura, E. Yoshida, K. Suzuki, Electron. Lett. 26, 1936 (1990)

[34] M. M. Choy, C. Y. Chen, M. Andrejco, M. Saifi, C. Lin, IEEE Photon. Tech. Lett.
2, 38 (1990)

[35] R. I. Laming, M. C. Farries, P. R. Morkel, L. Reekie, D. N. Payne, P. L. Scrivener,

F. Fontana, A. Righetti, Electron. Lett. 25, 12 (1989)

[36] J. E. Townsend, W. L. Barnes, K. P. Jedrzejewski, S. G. Grubb, Electron. Lett.
27, 1958 (1991) }

[37] S. G. Grubb, W. H. Humer, R. S. Cannon, S. W. Vendetta, K. L. Sweeney, P. A.
Leilabady, M. R. Keur, J. G. Kwasegroch, T. C. Munks, D. W. Anthon, Electron. Lett.
28, 1275 (1992)

[38] J. L. Zyskind, E. Desurvire, J. W. Sulhoff, D. J. DiGiovanni, JEEE Photon. Tech.
Lett. 2, 869 (1990)

[39] E. Desurvire, J. R. Simpson, Optics Letters 15, 547 (1990)

[40] M. Tachibana, R. I. Laming, P. R. Morkel, D. N. Payne, Optics Letters 16, 1499

(1991)

[41] J. W. Dawson, N. K. Park, K. J. Vahala, Appl. Phys. Lett. 60, 3090 (1992)

[42] J. R. Armitage, “Introduction of Glass Fiber Lasers and Amplifiers” in Optical Fiber
Lasers and Amplifiers editor P. W. France, CRC Press 1991

[43] R. I. Laming, L. Reekie, P. R. Morkel, D. N. Payne, Electron. Lett. 25, 455 (1989)
[44] K. Tajima, Electron. Lett. 26, 1498 (1990)

[45] Y. Kimura, M. Nakazawa, Electron. Lett. 28, 1420 (1992)

[46] R. P. Davey, K. Smith, A. McGuire, Electron. Lett. 28, 482 (1992)

[47] D. C. Hanna, A. Kazer, M. W. Phillips, D. P. Shepherd, P. J. Suni, Electron. Lett.
25, 95 (1989)

[48] F. Fontana, G. Grasso, N. Manfredina, M. Romagnoli, B. Daino, Electron. Lett. 28,
1291 (1992)

[49] A. Takada, H. Miyazawa, Electron. Lett. 26, 216 (1990)

[50] H. Takara, S. Kawanishi, M. Saruwatari, K. Noguchi, Electron. Lett. 28, 2095
(1992)

[51] X. Shan, D. Cleland, A. Ellis, Electron. Lett. 28, 182 (1992)

{[52] H. Takara, S. Kawanishi, M. Saruwatari, J. B. Schlager, Electron. Lett. 28, 2274
(1992)

[53] D. J. Richardson, R. I. Laming, D. N. Payne, V. Matsas, M. W. Phillips, Electron.
Lett. 27, 542 (1991)

[54] I. N. Duling, II, Electron. Lett. 27, 544 (1991)

[55] E. Yoshida, Y. Kimura, M. Nakazawa, Appl. Phys. Lett. 60, 932 (1992)

[56] N. J. Doran, D. Wood, Optics Letters 13, 56 (1991)

[57] D. J. Richardson, R. I. Laming, D. N. Payne, M. W. Phillips, V. J. Matsas,
Electron. Lett. 27, 730 (1991)

[58] N. Pandit, D. U. Noske, S. M. J. Kelly, J. R. Taylor, Electron. Lett. 28, 455 (1992)

[59] W. Lenth, R. M. MacFarlane, Optics and Phonics News, 8 (March 1992)

[60] J. Y. Allain, M. Monerie, H. Poignant, Electron. Lett. 27, 1156 (1991)

{61} J. Y. Allain, M. Monerie, H. Poignant, Electron. Lett. 26, 166 (1990)

[62] J. J. Whitley, C. A. Millar, R. Wyatt, M. C. Brierley, D. Szebestam, Electron. Lett.
27, 1785 (1991)

ERS

[63] R. G. Smart, D. C. Hanna, A. C. Tropper, S. T. Davey, S. F. Carter, D. Szebesta,
Electron. Lett. 27, 1307 (1991)

[64] A. Saissy, D. B. Ostrowsky, G. Maze, Applied Optics 30, 1933 (1991)

[65] E. W. J. L. Oomen, J. Luminescence 50, 317 (1992)

[66] G. A. Ball, W. W. Morey, W. H. Glenn, JEEE Photon. Tech. Lett. 3, 613 (1991)

[67] H. Sabert, A. Koch, R. Ulrich, Electron. Lett. 27, 2176 (1991)

[68] M. W. Maeda, J. S. Patel, D. A. Smith, C. Lin, M. A. Saifi, A. Von Lehman, JEEE
Photon. Tech. Lett. 2, 787 (1990)

[69] D. A. Smith, M. W. Maeda, J. J. Johnson, J. S. Patel, M. A. Saifi, A. Von Lehman,
Optics Letters 16, 387 (1991)

[70] P. R. Morkel, G. J. Cowle, D. N. Payne, Electron. Lett. 26, 632 (1990)

[71] K. Iwatsuki, H. Okamura, M. Saruwatarf, Electron. Lett. 26, 2033 (1990)

[72] G. J. Cowle, D. N. Payne, D. Ried, Electron. Lett. 27, 229 (1990)

[73] P. Barnsley, P. Urquhart, C. Millar, M. Brierley, J. Opt. Soc. Am. A 5, 1339 (1988)

[74] N. K. Park, J. W. Dawson, K. J. Vahala, C. M. Miller, Appl. Phys. Lett. 16, 150
(1991)

[75] S. L. Gilbert, Optics Letters 16, 150 (1991)

(76] J. L. Zyskind, V. Mizrahi, D. J. DiGiovanni, J. W. Sulhoff, Electron. Lett. 28, 1385
(1992)

[77] G. A. Ball, W. W. Morey, P. K. Cheo, to be published JEEE Photon. Tech. Lett.

[78] N. K. Park, J. W. Dawson, K. J. Vahala, Optics Letters 17, 1274 (1992)

[79] N. K. Park, J. W. Dawson, K. J. Vahala, to be published Optics Letters

[80] S. L. Gilbert, Proceedings Tenth International Conference on Laser Spectroscopy,
Font-Romeu, France, 17-21 June, 1991, eds. M. Ducloy, E. Giacobino, G. Camy, pp.
359-364

[81] J. Zhou, N. K. Park, J. W. Dawson, K. J. Vahala, M. A. Newkirk, U. Koren, B. I.

Miller, submitted to Appl. Phys. Lett.

[82] K. J. Vahala, N. K. Park, J. W. Dawson, S. Sanders, in AGARD Lecture Series
184, Advances in Fiber-Optic Technology in Communications and for Guidance and

Control, 1-1

Chapter 2

A Single-Frequency Fiber Laser

This chapter details the design, construction and testing of a single-frequency fiber
laser. The first part of the chapter discusses important design considerations such as the
choice of the resonator geometry, the erbium doped fiber amplifier and the constraints on
the tandem fiber Fabry-Perot filters which provide single frequency operation in
combination with broad band tunability. The second part of the chapter discusses the
details of all components in the laser, assembly of the laser and verification of the operating
parameters. Finally, the threshold lasing power, slope efficiency, output power, tuning
range, single-frequency operation and sidemode suppression are all experimentally

verified.

2.1 Design Considerations for Optimal Single-Frequency

Operation of an Erbium-Doped Fiber Laser

In order to achieve lasing, the first requirement is to have an optical gain equal to or
greater than the optical loss of the system. Erbium doped fiber amplifiers, which provide
greater than 40 dB of optical gain, are now commercially available. Amplifiers with gains
as high as 54 dB [1] have been demonstrated. With gains this large, improperly isolated

amplifiers will lase off their own cleaved end faces. Since oscillation is easy to achieve

using these devices, there exists a great opportunity to make high performance lasers.
Many different components can be inserted in the resonant cavity to perform a variety of
specialized functions. If a low loss component is not readily available, the amplifier will
easily compensate for a higher loss component allowing demonstration of a desired
concept. For example, in chapter 5, a phase modulator is inserted into the cavity to aid in
frequency stabilization.

The output power of a laser is dependent upon three major factors: the resonator
loss, Output coupling and the characteristics of the gain medium. In the case of the single
frequency fiber laser, the resonator loss can be estimated from the sum of the individual
component losses. A variable output coupler was not available at the time of the
experiment, so a fixed 50% output coupler was used. The only remaining parameters
necessary to estimate the laser output power are the gain characteristics of the erbium fiber
amplifier used in the laser.

Data on the gain characteristics are provided with commercial fiber amplifiers
(Figure 2.1). For research grade amplifiers this data must be measured in the lab. (While
some amplifiers were constructed during the course of this research, optimal EDFA design
has been discussed at length elsewhere [1, 2, 3] and that discussion will not be repeated
here.) The easiest way to do this is to use a fiber laser as a signal source. This necessitates
having two amplifiers. The experimental setup for measuring the gain characteristics is
shown in figure 2.2.

First the output of a fiber laser is collimated. The collimated beam is chopped,
attenuated and recoupled back into the fiber. The fiber gain responds slowly due to the 12
ms lifetime of the upper level. Chopping speeds above 200 Hz do not significantly
modulate the inversion in the amplifier, thus allowing lock-in detection of the amplified
signal. For small input signals (< 40 dBm in a commercial amplifier) the amplified

spontaneous emission power actually exceeds the amplified signal power. Lock-in

GAIN (68)
so

oe]

” “

GAIN (dB)
50.0

40.0

---8——-@-+—_9—-»-. |

30.0 oe

20.0

10.0

DO tee reg ura INPUT POWER (dBm)
. sees sor ear eee eee en cee Pree nceee sae renee
-50 -40 -30 -20 -10 0 10

OUTPUT POWER (dBm)
20

10 —-~i-@

ro INPUT POWER (dBm)
-50 -40 -30 +20 -10 0 10

Figure 2.1: Gain characteristics of a commercial EDFA.
Corning FiberGain, module P3-21, optical gain at 1532
nm, Al codoped Erbium fiber.

28
Chop. Fiber

= Coupler

Fiber Att.
Laser
50/50
LD
<— | PD
Lock-In #1
Lock-In #2
ISO.

— PD
/7 WDM \

Figure 2.2: Experimental set-up for measuring the gain
in an EDFA. WDM: wavelength division multiplexer,
ISO: optical isolator, PD: photo-diode, LD: pump laser,
Chop: chopper, Att: variable attenuator,

detection of the amplified signal allows it to be separated from the large background noise
and minimizes the error due to this noise at larger signal powers.

A fused fiber coupler acts as a beam splitter sending one-half of the signal directly
to a photo-detector and the other half through the amplifier to a second photo-detector.
Lock-in detection of the input and amplified signals gives an accurate measurement of the
gain. Gain may then be measured as a function of input power, output power, pump
power and wavelength (provided the fiber laser is tunable). Sample data is shown in figure
2.3.

Measurement of the small signal gain as a function of pump power, combined with
knowledge of the laser resonator losses, and the transmission of the output coupler yields
the pump power threshold for the laser. Once the gain has been plotted as a function of
output power (at maximum pump power), it is a simple matter to determine the amplifier
output power at which the gain is equal to the loss. The laser output power is the amplifier
output power at that gain reduced by the loss from the components between the amplifier
and the output coupler multiplied by the transmission of the output coupler. It should be
noted that for small resonator losses, simple formulas may be derived for laser output
power as a function of output coupler transmission [4]. However, the loss in a typical
fiber laser described in this thesis is greater than 10 dB and in this limit the assumption of
small losses does not hold.

Now that threshold and output power have been discussed, the requirements for
tuning and single frequency operation are considered. As mentioned earlier, the gain
bandwidth of an EDFA is nearly 50 nm; providing an appropriate tuning element can be
found, it is not difficult to achieve tuning over this entire range. Stable, single-frequency

operation is much more difficult to achieve, however.

30 T

Gain (a6)

+30 +25 +20 -18 10 +8 9
Output Power (dBm)

Galn (dB)

ati ee THEE Ea 9
“$0 -40 +30 -20 +10 °
input Power (dBm)

Output Power (dBm)

+20 |:o° bese {stierg diene

725 mt

wo UL
-$0 -40 +30 +20 “10 °
Input Power (dBm)

Figure 2.3: Measured gain characteristics of an EDFA.
12 m of York Erbium fiberEr concentration 100 ppm,
pump power 30 mW, wavelength 1537 nm, Ge-only
codoping.

If the optical field in the resonator is a standing wave, spatial holes will be burned
into the gain. The gain will not saturate at the nodes of the optical field, only at the
maxima. Other longitudinal modes of the resonator may be able to take advantage of these
unsaturated regions to achieve sufficient gain to oscillate. This effect was first observed by
Tang [5].

There are two common ways to achieve single frequency operation. One method is
to provide feedback to only one resonator mode within the gain bandwidth, by using a
short cavity and a very narrow band reflection filter [6]. The second method is to construct
a resonator using a ring geometry including optical isolators to ensure unidirectional
travelling wave operation. Spatial holes are eliminated in this case because the travelling
wave does not have nodes at fixed locations in space. The ring geometry also permits the
use of narrow band transmission filters as tuning and wavelength selection elements as
opposed to the narrow band reflection filters used in the linear cavity case.

Narrow band transmission filters such as fiber Fabry-Perot filters are commercially
available in an electronically tunable form. Furthermore, as more effort is being put into
realizing wavelength division multiplexed communications systems, more research
addresses electronically tunable narrowband transmission filters [7, 8]. These filters are
likely to improve in reliability, tuning range, bandwidth, and availability.

Narrow band reflection filters, by comparison, are not as well developed. There is
currently only one device of widespread interest: the intra-core Bragg grating
holographically written with an ultraviolet laser [9, 10]. This device is tuned by stretching
or heating the grating. The extent of its practical tuning range and speed is unclear.
Intracore Bragg gratings are also not yet commercially available.

For these reasons the fiber laser constructed here employs a unidirectional travelling
wave ring geometry with fiber Fabry-Perot filters as tuning elements. It is possible to
construct fiber Fabry-Perot filters with finesses as high as 2000 [11]. However, currently

available commercial devices typically have a finesse of about 120.

It is desirable for the laser's tunability to be limited only by the gain bandwidth and
not other components in the system. Therefore, the desirable free spectral range of the
fiber Fabry-Perot filter is at least the EDFA bandwidth. This is nominally 30 nm, but it is
possible to achieve laser operation over nearly 50 nm [12]. At a wavelength of 1550 nm,
50 nm corresponds to 6.244 THz. Assuming a finesse of 120 this gives a fiber Fabry-
Perot filter with a bandwidth of 52 GHz. Fiber lasers with narrowband transmission filters
other than a fiber Fabry-Perot but with transmission bandwidths close to this figure [12,
13, 14] have been observed to exhibit rapid mode-hopping over a range of several GHz.
To understand why this is so, it is necessary to consider the sidemode suppression of the
fiber laser.

The sidemode suppression in a single frequency laser is the ratio of the power in the
lasing mode to the power contained in the strongest longitudinal mode of the laser other
than the lasing mode. For stable single frequency operation, this ratio needs to be large. In
order to better understand the factors that govern the sidemode suppression, a brief
derivation of it will be presented at this point.

Consider the rate equations for the cavity modes. Assume the gain, g, is flat in the
region of interest. Let P| denote the power in the lasing mode and Pm denote the power
contained in a nearby longitudinal mode. Let 6 denote the spontaneous emission into a
longitudinal mode at a given gain, g. Finally, let Oy denote the loss in the lasing mode and

Qm, the loss in the sidemode of interest. Then the rate equations for the longitudinal

modes are

dP; _

dP,

——= £Py- OmP mt? (2.2)

The laser is assumed to be operating continuous wave, so the time derivatives in 2.1 and

2.2 may be set equal to zero to solve for the steady state power. Therefore,

P ° d P ° (2.3)

The sidemode suppression is defined to be the ratio of the power in the lasing mode to the

power in the sidemode in question. Therefore,

Pom 8 Om & (OG pn> % +)
5 = = 14 ~ (2.4)
m Oy - & a) - g ]

where equation 2.3 was used in the last step. The difference between the lasing mode loss

and the gain is very small. The loss difference between two modes should be a somewhat }

larger number if the laser is truly lasing in only one longitudinal mode. This justifies the

approximation in the last step of 2.4.
From 2.4 it is clear that the sidemode suppression increases as the loss difference

between the lasing mode and the nearest adjacent longitudinal modes increases. The

sidemode suppression also increases as the power in the lasing mode increases. The
amount of power in the lasing mode will be limited by the output power of the fiber
amplifier at the gain necessary to achieve oscillation. Decreasing the overall resonator loss
will increase the laser power and lower the amount of spontaneous emission into the mode,
thus increasing the sidemode suppression. Dramatic improvement is not possible using
this method, however.

This leaves the loss difference between the lasing mode and the adjacent sidemodes

to consider. The spacing of the longitudinal modes will greatly affect their loss difference.

The longitudinal mode spacing is inversely proportional to the length of the resonator. The
fiber laser's length must be at least as long as the EDFA plus length for optical isolators, an
output coupler, a polarization controller and the tuning filter. Using a normal EDFA this
puts a lower bound on the length of 25 m. Using an EDFA based on highly doped erbium
fiber, it is possible to reduce this length to around 4 m [15]. However, highly doped fiber
is neither commercially available or available in the preferred aluminium codoped, wide
bandwidth form. If one assumes a fiber laser length on the order of 25 m, the longitudinal
mode spacing of the resonator will be 8 MHz. As mentioned above, the tuning filter free
spectral range is 6.24 THz and the bandwidth is 52,000 MHz. Assuming a standard
Fabry-Perot transmission function, the loss difference between the mode directly under the
transmission peak and the mode that is 8 MHz away is 2.6X10-8. This is a very small
number and in practice is insufficient to provide good sidemode suppression.

It is possible to improve the loss difference by choosing a smaller free spectral
range and thus a smaller bandwidth. For example, a fiber Fabry-Perot filter with a free
spectral range of 10 GHz and a bandwidth of 125 MHz would give a loss difference of
1.6X10-2. This should result in an increase in sidemode suppression of 58 dB in
comparison to above. However, use of this filter alone would reduce the tuning range to
10 GHz.

To prevent the nearest neighbor longitudinal cavity modes from lasing and to
restore the wide tuning range, it is possible to combine the best aspects of the two fiber
Fabry-Perot filters by using them in tandem. A narrowband fiber Fabry-Perot filter (NB
FFP) with a bandwidth of 125 MHz could be employed to ensure a large sidemode
suppression and thus high stability. A broadband fiber Fabry-Perot filter (BB FFP) could
then be employed to select which NB FFP transmission peak had sufficiently low loss to
allow only one of the longitudinal modes of the resonator to lase. Figure 2.4 shows the
transfer functions of the two filters, the transfer function of the tandem pair with isolation

and the longitudinal mode spacing of the fiber laser.

The last important design consideration is optical isolation. The NB FFP and BB
FFP described above are constructed of two fibers with highly reflective dielectric mirrors
deposited on the ends. These mirrored fiber ends are brought close together to form an
optical cavity. With two FFP filters in the resonator, an optical cavity is also formed by the
space between the two filters. In order to ensure the tandem pair of filters is operating
correctly (i.e., the transfer function of the tandem pair is the product of the transfer
functions of the individual filters), an optical isolator must be placed between the filters.

Finally, isolation around the gain region must be considered. The loss difference
between adjacent longitudinal modes of the fiber laser is very small. Small amounts of
feedback to the very large gain in the EDFA can cause significant gain ripple. In short this
gain ripple can invalidate the assumption of a flat gain over the bandwidth of the NB FFP
used in the sidemode suppression calculation. (This is one reason the BB FFP is
insufficient to provide stable single frequency operation by itself.) Thus the sidemode
suppression can be significantly less than predicted. Furthermore, the gain medium length
can drift with temperature and the peaks and valleys of the gain ripple can move with
respect to the NP FFP filter peak, actually inducing mode hops.

On the surface one might think that this effect could be easily countered, however,
feedback as small as 1 part in 103 (30 dB of isolation at both ends of the EDFA) in
combination with a gain of 20 dB can lead to gain ripple induced mode hopping.
Fortunately, increasing the isolation to 40 dB or shorting the gain medium can eliminate the
problem. A general expression for the gain ripple is given in reference 16.

To summarize this section, there are several important points to consider when
constructing a single-frequency fiber laser. The first of these is ensuring low loss and
appropriate Output coupling based on that loss. Determination of the output coupling for
high loss lasers requires measurement of the saturation characteristics of the laser amplifier.
For wideband tuning combined with single-frequency operation, a narrowband

transmission filter in conjunction with a ring cavity configuration is found to give the best

\ re | i [\ FRR fl

L— (BB)
<)

{10

Figure 2.4: Tandem fiber Fabry-Perot filter concept.
h FP filter selects the coarse lasing wavelength.
FP provides sufficient loss difference between
longitudinal modes so that only one mode lases.

results. For stable operation, an NB FFP and BB FFP are combined in tandem to provide
broadband tuning in combination with high sidemode suppression. Finally optical isolation
must be considered. Isolation is required between the tandem fiber Fabry-Perot filters and
severe restraints are placed on the isolators surrounding the gain media due to the

detrimental effects of gain ripple on the sidemode suppression.

2.2 Construction of an All-Fiber, Widely-Tunable, Single-
Frequency Erbium-Doped Ring Laser

Figure 2.5 is a schematic of the basic fiber laser design that will be considered
throughout the rest of the thesis [17, 18]. The laser is a ring resonator with travelling wave
operation to prevent spatial hole burning. The gain element in the laser consists of a
commercial EDFA approximately 25 m in length. This EDFA is an aluminium codoped
module with a gain peak at 1532 nm and capable of providing 37 dB of optical gain at this
wavelength. The gain characteristics are shown in figure 2.1.

There are three isolators pictured in figure 2.5. These isolators are pigtailed with
single mode fiber and are designed to operate uniformly independent of the input
polarization state. The peak isolation is 40 dB at 1540 nm with a minimum isolation of 30
dB (from 1520-1560 nm) and an insertion loss of 0.6 dB. The isolators are placed as
discussed above. One isolator is used to inhibit interaction between the two tandem fiber
Fabry-Perot filters and one isolator is placed at each end of the EDFA to minimize the gain
ripple.

There are two fiber Fabry-Perot filters in the laser. These two filters act in tandem
to provide high sidemode suppression and stable single frequency operation in combination
with broadband tunability. The broadband fiber Fabry-Perot filter (BB FFP) has a free
spectral range of 4024 GHz or about 32 nm. The BB FFP’s finesse is 120 and the

corresponding bandwidth is 33 GHz. This filter provides broadband tunability for the

Figure 2.5: Erbium-doped fiber ring laser.FFP: fiber
Fabry-Perot filter, Pol: polarizer, PC: polarization

controller, k: output coupler, ISO: isolator, EDFA:
Erbium doped fiber amplifier

laser; its insertion loss is 3 dB. The BB FFP can be tuned over its entire free spectral range
by applying a voltage of 0-17 V to a piezo-electric stack bonded to the device.

The narrowband fiber Fabry-Perot filter (NB FFP) has a free spectral range of 10.2
GHz or about 0.1 nm. The NB FFP’s finesse is 80 and the corresponding bandwidth of
the device is 125 MHz. This filter provides high sidemode suppression and stable
operation for the laser; its insertion loss is 4 dB. It is not tunable. The cavity for the BB
FFP is an air gap between two closely spaced fibers whose facets are coated to form
dielectric mirrors. The NB FFP's cavity is longer than the BB FFP's cavity and so a piece
of fiber is inserted between the dielectric mirrors to prevent diffraction from lowering or
altogether eliminating the FFP finesse.

The intra-cavity fiber in the NB FFP must be fixed in place. The stresses that hold
this fiber induce birefringence in the NB FFP. As a result the NB FFP cavity length is a
function of the input polarization state and when inserted into the laser ring the laser
frequency will be a function of the laser polarization state. To prevent the laser polarization
state from fluctuating with temperature and other extrinsic effects, a polarizer and
polarization controller have been placed in the resonator.

The polarizer used in the cavity is a plasmon wave type polarizer. It is made from
polarization preserving fiber that has been lapped and polished close to the core along one
polarization axis. A metal is then deposited onto the fiber. Incoming light polarized
perpendicular to the metal coating is essentially unaffected. Incoming light polarized
parallel to the metal coating loses energy by exciting surface plasmon waves in the metal
coating. The polarization extinction ratio of this device is 27 dB and the insertion loss is
less than 0.5 dB (not counting losses in coupling to and from non-polarization maintaining
fiber to polarization maintaining fiber).

The polarization controler used in the laser consists of three fiber optic quarter
wave plates whose axes of birefringence can be rotated with respect to the others. Each

quarter wave plate is a two inch diameter disk about which a single mode optical fiber is

wrapped three times. Bending stresses from the fiber being wrapped around the disk
induce birefringence; three wraps provide the correct length for a quarter wave plate.

The output coupler of the laser is a 3 dB fused fiber coupler. These devices are
made by twisting two fibers about each other and fusing them together. The twisted
section is heated and stretched. During this process the cores of the two fibers are brought
close together and as a result there is coupling of light from one core to the other. By
monitoring light throughput during the stretching process, the degree of coupling may be
controlled. The twisted section is hermetically sealed in a rigid housing to protect it.
Wavelength division multiplexers that act as dichromatic mirrors or beamsplitters may be
also fabricated by this process.

All of these components are connected together either by fusion splices or by
reusable mechanical splices. A fusion splicer brings two cleaved fibers together using xyz
translational stages. An electric arc heats the fiber end faces as they are pushed together.
This technique can produce low loss (< 0.1 dB) low back-reflection (<60 dB) connections.
However, the splices require about 15-20 minutes to make in the lab and may have to be
done more than once to achieve good results (newer model fusion splicers can make splices
faster and more reliably).

A more convenient way to connect components together is to use mechanical
splices. These splices are reusable and can be made in about 2 minutes. They have an
insertion loss of 0.2 dB and a back-reflection of < 40 dB. A typical splice consists of a
guide to align the two fibers and index matching gel to avoid Fabry-Perot effects that might
occur due to the small air gap where the end faces that are brought together. A small piece
of plastic can be clamped around the jacketed regions of the two fibers to hold the splice
together. Due to convenience, in terms of allowing rapid assembly and interchangability of
components, mechanical splices are used for most connections in this research. Only

where very low back-reflection was required, such as between the EDFA and the isolators,

are fusion splices used.

The total cavity length of the laser for the configuration in figure 2.5 was 50 m
corresponding to a longitudinal mode spacing of 4 MHz. This length is somewhat longer
than the minimum achievable length. The extra length is primarily due to failure to shorten
the fiber pigtails on the components. The components are all relatively new products and
as such cost at least $2000 (with the exception of the fused fiber couplers). Shortening the
pigtails too much risks destroying the components. If components were to be dedicated to
fiber laser use, however, these pigtails could be cut and the components permanently
fusion spliced together to form a minimum cavity length device.

The total loss in the laser including the output coupler is 15.3 dB. From figure 2.1
the expected output power of the EDFA at this gain is approximately +10 dBm. This
power is attenuated by a 9.0 dB loss between the amplifier output and a photodiode outside
the cavity, so that a maximum output power of +1 dBm or 1.25 mW is expected. If the BB
FFP were moved to the other side of the output coupler, the output power would increase
by 3 dB to approximately 2.5 mW output power.

Looking at figure 2.1 one might be led to believe that a higher output coupling
would increase the output power. This is true at the gain peak of 1532 nm; however, a
look at gain saturation data at other wavelengths quickly indicates that some other
wavelengths would not lase at higher output coupling/higher cavity loss. A 3 dB output
coupler gives a relatively flat maximum output power as a function of wavelength, since

most wavelengths can readily achieve 15 dB of gain.

2.3 Lasing Characteristics of an All-Fiber, Widely-Tunable,
Single-Frequency Erbium-Doped Ring Laser

The laser described above was constructed and fully tested. In order to verify that
only one longitudinal mode of the cavity was lasing, the output spectrum was examined

with both a spectrometer and a Newport Research Super Cavity Fabry-Perot. The

spectrometer was a SPEX model 340E spectrometer with a resolution of 0.15 nm which
corresponds to 18.7 GHz. A spectrometer scan over the full bandwidth of the EDFA
(1500-1600 nm, to be absolutely sure) showed only one lasing peak.

The Super Cavity has a free spectral range of 6 GHz and a finesse of greater than
6000. The bandwidth is 1 MHz which is sufficient to resolve the 4 MHz mode spacing of
the fiber laser. A scan over one free spectral range of the cavity shows only one lasing
mode (see figure 2.6). The smaller peaks in figure 2.6 are transverse modes of the Fabry-
Perot analyzer excited by imperfect alignment of the input beam with the fundamental mode
of the Super Cavity. These peaks are identifiable by their 800 MHz spacing from the main
peak. Figure 2.7 provides a closer look at the lasing mode. The scale in figure 2.7 is 12.5
MHz per division (125 MHz total). This scale is sufficiently narrow so that it may be
clearly observed that no other modes are lasing.

Observing the scan from the Super Cavity over a long period of time, the lasing
mode can be observed to drift to the left or right, due to thermal expansion/contraction of
the laser cavity. As the cavity length expands 1.0 um (one wavelength in glass), the lasing
mode moves out from under the NB FPP transmission peak and one of the nearest
longitudinal cavity modes moves under the transmission peak. At this point a mode hop of
4 MHz occurs. This occurs every 30 seconds to a minute. In comparison to previous
single frequency fiber ring lasers [12, 13, 14], this is a great improvement since all were
observed to mode hop rapidly over several GHz.

To verify the tuning range, the voltage across the BB FFP filter was increased from
0 to 17 VDC in 1 V increments (sufficient voltage to tune the BB FFP one free spectral
range). The wavelength was then measured using the spectrometer and single-frequency
operation was simultaneously monitored with the Super Cavity. Laser wavelength as a
function of voltage applied to the BB FFP is shown in the graph in figure 2.8. (The NB
FFP has been moved to the opposite side of the output coupler so the laser is configured

for maximum power output.) The laser tunes in 10 GHz jumps from 1528-1558 nm. Two

Figure 2.6: Output spectrum of the fiber laser as seen
with a Newport Research Super Cavity. The free
spectral range is 6 GHz. The smaller peaks are
transverse modes of the Super Cavity.

Figure 2.7: Output spectrum of the fiber laser as seen
with a Newport Research Super Cavity. The scale is
12.5 MHz/div. The linewidth is limited by the resolution
of the supercavity, which is 1 MHz.

modes lase at 11 V, as the gain at 1559 nm approximately equals the gain at 1527 nm
(wavelengths that are one BB FFP free spectral range apart).

The laser output power, threshold and slope efficiency were also measured. A
typical plot of output power vs. pump power is shown in figure 2.9. The threshold power
is around 15 mW. The maximum output power is 1.2 mW as expected and the slope
efficiency is approximately 9%. Figure 2.10 shows the output power as a function of
wavelength. The output power vs. wavelength is reasonably flat as expected, since for
EDFA operation at about 15 dB gain, the output power is approximately + 10 dBm ( +/- 1
dB) independent of the operating wavelength.

To complete the initial characterization of the laser properties, the sidemode
Suppression was also measured. For this measurement the laser output was directed onto a
photodiode. A low frequency bias T (10 kHz - 18 GHz) separated the DC component from
the AC component of the resulting photo-current. The DC component is measured with an
ammeter. The AC component is amplified and sent to an RF spectrum analyzer, where the
beating between the sidemodes and the main lasing mode can be observed. The DC photo-
current was used to obtain the power in the main mode. The power due to the beating
between the main mode and the nearest sidemodes was then read from the spectrum
analyzer. From these values the sidemode suppression (ratio of the power in the lasing
mode to the power in the largest sidemode) of the laser may be determined. This ratio is as
large as 60 dB, which is a good value (for telecommunications systems, 30 dB sidemode
suppression is the benchmark for digital communications systems and 60 dB for analog
communications systems). The sidemode suppression as a function of laser wavelength is

plotted in figure 2.11.

1560 , mT , ee T Tt T 1 a ee T T T

Lprpipy it

1555 | 5

1550 { ©

1545

[oe ie ee |

1540 fmm

.6)

1535 F >

Wavelength (nm)

1530

Depiri toys tts

1525

1520 ; 4 : 1. L 1 t 1 4 tt
0 5 10 15 20

BB FFP Voltage (V)

Figure 2.8: Wavelength as a function of voltage applied
to the BB FFP. The break at 11 V is the point at which
the gain at 1558 nm approximately equals the gain

at 1528 nm. The spacing between these wavelengths is
one free spectral range of the BB FFP.

nm
XY
re)
(e)
oO

Via ,

Vu
ana :

0 10 20 30 40

Laser Output Power (mW)
“ in

Pump Power (mW)

Figure 2.9: Output power vs. pump power. Wavelength:

1550 nm, threshold: 7 mW, slope efficiency: 9%. The
roll off at high power is the pump diode lasing spectrum

shifting away from 980 nm, changing the pump power
absorption.

Seaman

4 oc 1

3.5 | ‘

F 5 ? ae 4

=~ 3°CéC€#é€, o O 4

= C 2. Qe Quenectes 7

Cdl = =

ae} : 4

2 = =

5S 1.5 +f :

Oo F 1

o 4

7) 1 1

wo 4

§ ; :

0.5 ¢ ‘
ot

1525 1530 1535 1540 1545 1550 1555 1560 }
Wavelength (nm)

Figure 2.10: Output power vs. wavelength.

Sidemode Suppression (dB)

Figure 2.11: Sidemode suppresion vs. wavelength.

e) OQ 6 O 050° 0O..0-0--© One)
ib 4
1530 1537 1545 1552 1560

Wavelength (nm)

Properties of the laser intensity noise including a more detailed discussion of the

sidemode suppression will be presented in the next chapter.

References

[1] R. I. Laming, M. N. Zervas, D. N. Payne, JEEE Photon. Tech. Lett. 4, 1345 (1992)

[2] E. Desurvire, J. R. Simpson, J. Lightwave Tech. 7, 835 (1989) ’

[3] B. Pedersen, A. Bjarkev, J. H. Polvsen, K. Dybdal, C. C. Larsen, J. Lightwave Tech.
9, 1105 (1991)

[4] Amnon Yariv, /ntroduction to Optical Electronics, Third Ed. (Holt, Rinehart and

Winston, New York, 1985) p. 160

[5] C. L. Tang, H. Statz, G. de Mars, J. Appl. Phys. 34, 2289 (1963)

[6] G. A. Ball, W. W. Morey, W. H. Glenn, /EEE Photon. Tech. Lett. 3, 613 (1991)

[7] D. A. Smith, paper FB1 in OFC/IOOC 1993 Technical Digest Series Volume 4 (Optical
Society of America), 244 (1993)

[8] Z. Tang, H. F. Taylor, O. Eknoyan, V. P. Swenson, paper FB2 in OFC/IOOC 1993

Technical Digest Series Volume 4 (Optical Society of America), 245 (1993)

[9] G. Meltz, W. W. Morey, W. H. Glenn, Optics Letters 14, 823 (1989)

[10] R. M. Atkins, V. Mizrahi, Electron. Lett. 28, 1743 (1992)

[11] J. Stone, L. W. Stulz, Electron. Lett. 27, 2239 (1991)

[12] M. W. Maeda, J. S. Patel, D. A. Smith, C. Lin, M. A. Saifi, A. von Lehman, JEEE
Photon. Tech. Lett. 2, 787 (1990)

[13] K. Iwatsuki, H. Okamura, M. Sarvwatary, Electron. Lett. 26, 2033 (1990)

[14] D. A. Smith, M. W. Maeda, J. J. Johnson, J. S. Patel, M. A. Saifi, A. von Lehman,
Optics Letters 16, 387 (1991)

[15] J. L. Zyskind, J. W. Sulhoff, et al., Electron. Lett. 27, 2148 (1991)

[16] T. Saitoh, T. Mukai, /EEE J. Quantum Electron. 23, 1010 (1987)

[17] N. K. Park, J. W. Dawson, K. J. Vahala, C. Miller, Appl. Phys. Lett. 59, 2369
(1991)

Sea

Pa:

——s

[18] J. W. Dawson, N. K. Park, K. J. Vahala, C. Miller, paper WAS in in Optical Fiber
Communication 1992 Technical Digest Series Volume 5 (Optical Society of America), 99

(1992)

Chapter 3

Intensity Noise Characteristics of a Single-
Frequency Fiber Laser

This chapter investigates the intensity noise characteristics of the single-frequency
fiber laser discussed in this thesis. The spectral content of the laser noise is the focus of the
investigation. The intensity noise spectrum consists of three distinct spectral regions. The
first region is defined by the free spectral range of the laser cavity. This region extends
from DC to one half the free spectral range of the laser cavity. The second and third
regions are defined by the bandwidth of the NB FFP. The second region consisting of
frequency span from the end of the first region to roughly twice the NB FFP bandwidth,
followed by the third region.

In the third region the intensity noise is very low. In order to accurately measure it
a balanced homodyne system is employed. This system allows one to make a calibrated
measurement of the intensity noise relative to the shot noise limit. It is shown that by
proper configuration of the laser cavity, it is possible to reduce the excess noise in this
region to the shot noise limit.

Each of these regions is discussed in detail in terms of accurate measurement of the

noise power per unit frequency. The measurement systems used to determine the noise

power are also discussed. A brief discussion of the expected noise spectra is included to

provide perspective.

3.1 Low and Intermediate Frequency Intensity Noise Spectra

As mentioned above the intensity noise spectra can be broken down into three
distinct spectral regions. The first of these is the region from DC to one-half the free
spectral range of the laser cavity. For the laser discussed in this thesis, this is DC to 2
MHz. At these low frequencies, one expects the primary source of intensity noise to be
shot noise driven relaxation oscillations.

Relaxation oscillations in a laser occur due to interaction between the lasing field
and the inversion in the optical gain medium [1]. To see how this works, consider the
consequences of a small fluctuation in the intensity of the laser field in the cavity. For
example, if there is a small decrease in the lasing field, fewer emissions are stimulated from
the optical gain media. Assuming a constant pumping rate, this results in an increase in the
inversion, which leads to an increase in the optical gain. The increase in the optical gain
results in an increase in the optical field intensity.

Thus gain saturation acts as a restoring force for laser field fluctuations. It can be
shown by a small signal analysis of the laser rate equations that the interaction between the
optical gain and the laser field has the mathematical form of a damped driven simple
harmonic oscillator [1]. The driving force for the system is spontaneous emission from
the gain medium. The characteristic times of the interaction are the photon lifetime in the
passive cavity and the spontaneous lifetime of the gain media.

The interaction has a natural resonance frequency known as the relaxation

oscillation resonance frequency, Wp. This frequency is given by the formula [1},

op = (-1) fir = A] &) (3.1)
tot \27 tot

where tc is the photon lifetime in the cavity, T is the spontaneous lifetime of the gain
medium and r is the ratio of the pump power to the threshold pump power. The
approximation in 3.1 is justified in this case since the photon lifetime of the fiber laser is
approximately 0.25 us, the spontaneous lifetime of the gain medium is 12 ms, and r = 5 at
maximum power. Equation 3.1 yields an expected relaxation oscillation frequency of 36.5
kHz for these values. Further from equation 3.1, it is expected that the relaxation
oscillation frequency will increase as the square root of the optical power (which is
proportional to r - 1).

In measuring the intensity noise spectra in the low frequency regime, a direct
detection system in combination with an RF spectrum analyzer was used. The noise power
is characterized in terms of the relative intensity noise. For a directly detected field, the
relative intensity noise (RIN) is defined as the ratio of the mean Square power per unit
bandwidth of the fluctuating photocurrent to the average photocurrent power. It is possible
to show [2] that for the case of spontaneous emission driven relaxation oscillation noise,

the RIN as a function of frequency, QQ, is given by

= + a
™R
RIN = 2 A® op (3.2)
2 2.2
(Mp - QU )~ + 2

where Tp is the stimulated lifetime of the gain medium and A@cy is the Schawlow-

Townes linewidth.

If the dominant noise source in the system is relaxation oscillation noise, the
measured RIN should have a frequency dependence similar to that described by equation

3.2. For typical parameter values, equation 3.2 has a peak, which is located approximately

at Op. For QQ << Wp, 1/TR the RIN is relatively flat as a function of frequency. For Q
>> Wp, 1/TR the RIN rolls off as 1/02.

To measure the relative intensity noise, the direct detection system shown in figure
3.1 was employed. The output of the fiber laser is collimated using a fiber input-output
coupler, then focused onto a photodiode. The photodiode used was a BT+D model
PDH0004 made of InGaAs/InP with a 50 micron diameter active region. It has a
responsivity of 1.0 A/W at 1550 nm and a bandwidth of 2.4 GHz. The photodiode is
connected to a bias T and reverse biased to 10 V. A current meter is connected in series
with the DC bias to monitor the average photocurrent. In addition an electrical RC low
pass filter with a 1 kHz corner frequency was connected to the DC port of the bias T to
eliminate low frequency electrical noise from the power supply line.

The bias T is a Picosecond Pulse Labs device model 5575A with a 32 ps rise time,
a 10 kHz low frequency cutoff (which set the low frequency cutoff for the measurement),
and a 0.6 dB insertion loss. The AC port of the bias T is connected to a Trontech RF
amplifier model W500K in series with another RF amplifier (same model) and then to a
Tektronics RF spectrum analyzer model 2782. Each of the RF amplifiers has a gain
bandwidth from 1 kHz to 500 MHz, a gain of 30 dB, and a noise figure of 2.2 dB.

Calibration of the system was accomplished by using an HP model 8640B signal
generator. The output power of the signal generator was adjusted so that the amplifier
system had the same thermal noise floor during the calibration as it had during the
measurement. At high frequencies (>300 MHz) the laser intensity noise power sunk below
the thermal noise floor of the measurement system. Verifying that the thermal noise floor
of the measurement system was the same during the measurement and calibration in these

frequencies permitted gain saturation of the amplifier to be removed as a source of

Spectrum
(BB) (NB) | T >on Analyzer

AMP
PC
POL PD

Figure 3.1: Experimental] set-up for measuring the low
frequency intensity noise spectra. PC: polarization
controller, POL: polarizer, DC: power supply, I: current
meter, T: bias T, AMP: RF amplifier, PD: photo-diode,

K: output coupling coefficient.

systematic error in the measurement. The oscillator frequency was varied over the
measurement range and the system gain was calibrated as a function of frequency. The
tandem amplifiers were found to have a gain of 48 dB by this technique. The gain variation
over the frequency range from 10 kHz to 500 MHz is a maximum of +/- 1.0 dB. As these
frequencies are much lower than the photodiode bandwidth, the frequency response of the
photodiode was assumed to be flat over this range.

During the measurement the RF spectrum analyzer is used to average 10
consecutive spectrum analyzer traces in order to provide a more accurate reading of the
noise power. The data is then read into a 486 DX personal computer using software from
Tektronix. It is possible, using the computer, to divide out the average photocurrent power
and the resolution bandwidth, converting the data to RIN. In addition the calibration data is
used to correct the RIN for the net system gain between the output of the laser and the RF
spectrum analyzer. The resulting calibrated data is printed out for detailed study.

A typical RIN spectrum in the frequency region from 10 kHz to 1 MHz is plotted in
figure 3.2. The resolution bandwidth of the spectrum analyzer for this trace was 10 kHz
and the output power of the laser was 813 UW. This output power corresponds to r = 4.6
for this configuration of the laser. In figure 3.2 the RIN of the laser is observed to increase
steadily from about 20 kHz to a peak of -105 dBc/Hz at 32 kHz (confirmed by decreasing
the span and resolution bandwidth of the spectrum analyzer to zoom in on the peak) and
then roll off at 20 dB/decade above 100 kHz. The expected relaxation oscillation frequency
from equation 3.1 is 35 kHz in good agreement with the measured value.

In order to positively identify the RIN spectra as shot noise driven relaxation

oscillations, equation 3.2 was used to fit the data in figure 3.3. The measured value of Op

was used as well as the theoretical value of A@oT, 0.003 Hz. TR was used as the fitting
parameter and was determined from the fit to be 0.1 ms. In figure 3.3, it can be seen that
the measured data at low frequency does not agree well with the theoretical fit. It is

suspected that a source other than spontaneous emission is driving the system at low

~N -95

~~ q

Oo q|

S$ -105])

ell)

> |

@ —125:

o —1354 VA, ny E
2 NSM stelabahi\atddad aki )
3 —145 | | | Hees
ce 0 200 400 600 800 1000

Frequency (kHz)

Figure 3.2: Relative intensity noise spectrum of the
fiber laser from 10 kHz to 1 MHz.

— -1007

N 4

= 4054

oO 1

a 110°

4 MS5

‘Oo -1204

s 1254

@ -130-

e -1354

rab) -1404

2 %: |
7% -1454 ‘
2 1504

10 100 1000

Frequency (kHz)

Figure 3.3: Theoretical fit of figure 3.2 data to equation
3.2. The discrepancy at low frequency is probably due
to a noise source other than spontaneous emission
driving the relaxation oscillation.

frequency. One possibility is room microphonics interacting with the long cavity length to
create length fluctuations or time dependent birefringence.

As a final check that the source of the low frequency noise is indeed relaxation
oscillation noise, the relaxation oscillation frequency is plotted as a function of the laser
output power in figure 3.4. The value of the pumping parameter f was determined from
the output power vs. pump power curve. The theoretical line going through the data points
in figure 3.4 was determined by plotting equation 3.1 as a function of power based on the
measured relation between r and power and using a tc of 0.25 us and T of 12 ms,

As described above, the intensity noise of the laser rolls off at 20 dB per decade of
frequency from 100 kHz out to 2 MHz. At this point the intensity noise from the two
sidemodes adjacent to the lasing mode begin to cause an increase in the laser intensity
noise. This effect can be seen in figure 3.5. Figure 3.5 is a spectrum analyzer trace of the
frequency range from 1 MHz to 20 MHz taken using the same experimental setup described
above for measuring the relaxation oscillation noise in the frequency region below 1 MHz.
The average laser power ( 813 LW) has been divided out, but the resolution bandwidth (30
kHz, increasing the resolution bandwidth did not change the peak heights) was not divided
out of the relative noise power.

The dominant intensity noise source in this frequency regime is clearly the main
lasing mode beating with amplified spontaneous emission in the adjacent longitudinal
modes of the resonator. The adjacent longitudinal modes do not have sufficient gain to
lase; however, the net loss for these modes is still quite small, hence their linewidths are
quite narrow. Because of the narrow bandwidth peaks in the intensity noise spectrum, it is
not possible to characterize the noise in this region as broadband noise, so units of dBc/Hz
are not used. In this frequency region sidemode suppression (dBc) is the best indicator of
the laser intensity noise. The theoretical factors governing sidemode suppression have

been discussed in section 2.1.

30 ri l r i ‘ 1 1 { l ! 1 | ! { L i i {4 i \ i
N 1 i
> 4 L
a 1 fe) fi
S 20 7 O L
o 4 i
® J t
LL ;
6 155 9 7
8 10 - L
oO 4 i
c 4 [
E54 :
o 1 J
Ss |
o 1 r |
0 a Tt ;
0 0.2 0.4 0.6 0.8 1 1.2

Output Power (mW)

Figure 3.4: Relaxation oscillation frequency vs,fiber
laser output power. The data has been fit to equation 3.1.

~ ~3s/ :

cD 30 kHz res. BW }

2 57) :

Oo 4

< |

> 7874 |

® | | ' t

5 77P |

ei FOUR UI

Pook A A A A

Frequency (MHz)

eee

Figure 3.5: Intensity noise spectrum of the fiber laser
from 1 MHz to 20 MHz. The sidemode suppression of
the laser is at least 60 dB. The laser output power was

813 LW.

It can be seen in figure 3.5 that the third sidemode at 11 MHz is the largest. Its
sidemode suppression relative to the main lasing mode is 62 dB. Figure 3.6 shows a
spectrum analyzer trace at the same power level with a resolution bandwidth of 300 kHz
only looking at the overall trend in the sidemode suppression from 1 MHz to 400 MHz. A
pattern can be seen; approximately every third peak is larger than its nearby neighbors.

The frequency spacing of these peaks is approximately 11-12 MHz. If the
amplified spontaneous emission spectra was to experience a frequency dependent loss or
gain from, for example, a parasitic Fabry-Perot cavity, an 11-12 MHz trend would indicate
a Fabry-Perot cavity about 9 m in length. A cavity of this length is most likely occurring
between one of the high reflectivity dielectric mirrors in the BB FFP and a splice in the
laser. Most probably the problem splice is at the interface between the polarization
preserving fiber attached to the polarizer and the normal fiber that makes up the rest of the
cavity. This problem could be eliminated in the future by adding another isolator to the
cavity or by making the entire cavity out of polarization preserving fiber.

Figure 3.6 clearly shows the sidemode suppression rolling off as a function of
frequency in the general trend expected based on equation 2.4. At around 100 MHz, it
begins to become difficult to distinguish the individual longitudinal modes. At around
twice the NB FFP filter bandwidth or about 250 to 300 MHz, the peaks disappear almost
completely. As shown below beyond these frequencies, it is again possible to characterize
the intensity noise spectra as broadband noise. However, for these frequencies the noise
level is quite close to the thermal noise floor of the measurement system. In order to
accurately characterize the intensity noise in this higher frequency, lower noise regime, a

better measurement system must be constructed. This is discussed in detail in the next

section.

~ 790
fs , 300 kHz res. BW :
xe) ;
‘804
= :
< -70
c |
Z al m |
= eal TENE oie
3 | . (Aut
= _100! | iy VENA tg
0 100 200 300 400

Frequency (MHz)

Figure 3.6: Intensity noise spectrum of the fiber laser
from 1 MHz to 400 MHz. The laser output power

was $13 uW.

3.2 Measurement of the High Frequency Intensity Noise Spectra
Using the Balanced Homodyne Technique

The intensity noise of a laser at a given frequency is shot noise limited if the noise
power at that frequency is due to random arrival of photons at the input of the detection
system. This random arrival is governed by Poissonian statistics and has a power spectral
density of 2eI, where e is the charge of an electron and I is the average photocurrent at the
photodetector.

Uncertainty relations (between photon number and phase, for example) guarantee
that any physical system has a minimum amount of noise. When the uncertainty in a
system is solely due to the uncertainty relation and this uncertainty is equally divided
between the two observables of the relation, the system is said to be in a minimum
uncertainty state [3]. In a minimum uncertainty state the uncertainty of each of the two
observables is said to be at the standard quantum limit (SQL). If the intensity noise of a
laser is shot noise limited, the intensity noise is at the SQL for the system.

This is very interesting because it implies the system is approaching a fundamental
operating limit. (Uncertainty in the observable can be reduced below the SQL but only at
the expense of an increase in the uncertainty in the conjugate observable [4, 5, 6].)
Measurement of the phase noise would be required in order to claim the system was in a
minimum uncertainty state. However, intensity noise at the standard quantum limit in
combination with single frequency laser operation is an excellent start towards this goal.
Furthermore, the low intensity noise properties of such a laser would have many
applications in measurements where low laser intensity noise is an important parameter,
such as some fiber optic sensor systems.

The standard technique for measuring the intensity noise of a laser was discussed in

section 3.1. However, this technique does not permit accurate determination of the

standard quantum limit. The balanced homodyne system [7] (figure 3.7) was developed to
allow accurate measurement of the SQL. The balanced homodyne system measures the
SQL when the two amplified photocurrents from detectors D1 and D2 are coherently
subtracted and the laser noise power when the two photocurrents are coherently summed in
a hybrid junction.

The two detectors D1 and D2 are the same model as the detector used earlier. The
amplifiers are Avantek amplifiers model ACT10-213-1, frequency range 10-1000 MHz,
gain 52 dB (+/- 1.5 dB) and a noise figure of 3.7 dB. The RF spectrum analyzer (same as
above) is operated in the zero span mode. In this mode the noise power within one
resolution bandwidth of the set frequency is continuously measured. The chopper and
lock-in amplifier (EG&G, model 5208) distinguish the laser noise power from the thermal
noise of the measurement system, thus increasing the sensitivity of the measurement. The
chopper modulates the signal beam at around 200 Hz which will not affect measurement of
the noise power at several hundred MHz. The microwave mixer is used to add or subtract

, of

the photocurrents.

The path lengths between the beamsplitter and the mixer must be carefully balanced
so that in the subtract mode the mixer eliminates all of the excess noise from the laser.
Balancing is performed by injecting light from a 1.5 um distributed feedback
semiconductor laser into the unused port of the fiber laser output coupler and modulating
the light at the frequency at which one wishes to measure the intensity noise. The optical
intensities reaching the two detectors are adjusted by rotating a half-wave plate before a
polarization sensitive beamsplitter, and the path lengths of the two arms are adjusted to give
a common-mode rejection of > 50 dB. A polarizer before the balanced homodyne system
maintains the intensity balance when the fiber laser output is measured and an external
polarization controller is optimized before each measurement to ensure that maximum
power is transmitted to the measurement system. The shot noise floor is confirmed by

retracing it with the laser output power fixed and externally attenuated.

X/A

D1 Al

laser Pol
output |
chopper
Lock-in
Amplifier

Tek 2782
Spectrum
Analyzer

Figure 3.7 Balanced homodyne detection system to

determine laser noise power relative to the SQL (shot

noise floor). The SQL is measured when the

photocurrents are differenced; the laser noise is measured
when the photocurrents are summed. (HWP: half-wave

plate; D1 and D2: photo-detectors; Al and A2: RF

amplifiers),

eae ee
Sa aetna
peer

A correction for losses of 56% between the output of the laser ring and the detectors

is given by:

Wobs =nwil-n (3.3)

where Wop, is the observed noise power normalized to the SQL; 1) is the overall
efficiency, and W is the actual noise power normalized to the SQL. As actual calibration of
the noise power throughout the entire measurement system is difficult and probably not that
accurate, the noise power data will be presented in either arbitrary units or normalized to the
SQL.

The noise power versus laser power at a fixed wavelength is shown in figure 3.8,
along with the shot noise limit. The laser configuration is shown in figure 3.1 (later this
configuration will be altered). The noise power varies linearly with the output power.
Because of the linear dependence of noise power on laser output power, the noise can be
characterized in terms of noise power relative to the SQL by the ratio of the slopes of the
two lines. Figure 3.9 shows the noise power relative to the SQL as a function of RF
spectrum analyzer frequency. A series of resonances with a 4 MHz period are observed,
which are clearly very weak sidemodes of the laser. The combination of the still visible
sidemodes andthe linear increase in noise power with laser power indicates that the
measured noise arises from beating between the lasing mode and amplified spontaneous
emission in the strongly suppressed sidemodes. The amplified spontaneous emission
power is independent of the amount of power in the lasing mode because the inversion of
the EDFA is clamped at the threshold inversion. Thus as observed, the beat noise between
the lasing mode and the amplified spontaneous emission will increase linearly with the
power in the lasing mode.

The noise power data in figure 3.9 were taken at around 300 MHz in frequency. If

the dominant noise source is beating noise between the amplified spontaneous emission in

154 “. aaa 1 an wan

Noise power (a. u.)

100 200 300 400. 500
Laser power (uW)

Figure 3.8: Noise power (arbitrary units) vs. laser power.
50% output coupling, 300 kHz resolution bandwidth.
Top line: balanced homodyne in sum mode; Bottom line:
balanced homodyne in difference mode.

30 1 i F aI 4 | 4 | ee ee | nm I 1 }
oO 1 .
©S 95 - NB FFP before gain -
a | 4
o 7 oo ©
2 20 - ° ° ° 5 —
oO 7 f 6 ° So © 96
ab) “4
2 15-7 -
& J
@ 4
6 10 - P
= .
(@) 4
QoQ. 4
3 54 7
5 4
z 7 f
0 }— a se ee es

306 307 308 309 310 311 312 313 314
Frequency (MHz)

Figure 3.9: Noise power relative to the SQL as a function
of spectrum analyzer frequency. 50% output coupling,
300 kHz resolution bandwidth.

the sidemodes and the lasing mode it should be possible to reduce this noise by using the
NB FFP filter, which has a bandwidth of 125 MHz, to selectively attenuate the sidemodes
and not the lasing mode. The data in figure 3.9 were taken with the NB FFP filter before
the gain and the BB FFP filter after the gain. The BB FFP has a bandwidth of 33 GHz
resulting in virtually no attenuation of the sidemodes at 300 MHz. The NB FFP filter has
an attenuation of 17 dB at 300 MHz away from its transmission peak. Therefore, if the
position within the cavity of the NB FFP and BB FFP are exchanged, there should be a 17
dB reduction in the intensity noise of the laser at this frequency.

Figure 3.10 shows the noise power relative to the SQL for the case of the NB FFP
filter after the gain as in figure 2.5 and for the case of the NB FFP filter before the gain as
in figure 3.1. A reduction of the intensity noise by 15 dB is apparent. Further, the noise
power relative to the SQL is now < 5 dB, which is quite close to optimal performance. To
further decrease the noise power relative to the SQL, the effect of the output coupler on the
noise characteristics was investigated.

The transmission of the output coupler affects the intensity noise in two ways.
First, shot noise is directly proportional to the laser output power. Because of the large
loss in this laser, the output power increases roughly as the transmission of the output
coupler increases, until the gain is no longer fully saturated (see section 2.1, 2.2). Thus
the SQL will increase linearly with the output coupler transmission. The amount of
amplified spontaneous emission from the gain medium coupled out of the laser will also
increase linearly with increasing output coupling. Therefore, the noise power due to beat
noise at the photodetector will increase as the square of the output coupler transmission
while the SQL increases linearly with the output coupler transmission. If the output
coupling decreases, then the ratio of the noise power relative to the SQL will decrease
linearly with it.

Furthermore, decreasing the output coupler transmission coefficient will decrease

the amount of gain required to achieve oscillation, which will decrease the amount of

spontaneous emission from the laser. The amplified spontaneous emission power has a

spectral density given by, Ngp(G-1 )hv, where G is the power gain of the amplifier, n is

the optical frequency and Nsp is the integrated spontaneous emission factor given in Ref. 8

by
—— G - Zz ' 1
np > | dz a(2) mgg(z) Jy 2 G4)

where 9(Z) = O,No(z) - O,N,(z) and gp = O,No(Z)/g(Z); Og and GO, are the

emission and absorption cross sections, respectively, N, and Nz are the upper and lower
state population densities, respectively, and | is the total amplifier length. (When Ngp is
independent of Z,Ngp = Ngp .) Clearly, the dependence of the spectral density of the
amplified spontaneous emission power on the gain is not linear when the spontaneous

emission factor is a function of position in the amplifier. However, there is still a general

trend of decrease in amplified spontaneous emission for a decrease in gain. }

Three output couplers were available for testing this concept. The output couplings

available from these couplers were 10%, 50% and 90% transmission. The 90% output

coupler resulted in decreased tuning range over the 50% output coupler because it increased

the gain necessary to achieve oscillation. The data from the 50% output coupler has already
been presented in figure 3.10. The data from the 90% coupler is presented in figure 3.11
and the data from the 10% output coupler is presented in figure 3.12.

Again the placement of the NB FFP filter is seen to have a dramatic effect on the

noise power relative to the SQL. The transmission of the 90% output coupler is 2.55 dB

higher than that of the 50% output coupler. Comparison of figures 3.10 and 3.11 show the
relative noise power increased by about 5 dB, 2.55 dB from the increased output coupling

and the rest from an increase in amplified spontaneous emission due to the increased gain

30
— 25
—!

CG
2)
2 20
@®
2 15
g 10
oO
eo.
B® 5
Oo
za

Figure

| = | i 4 4 1 { i]
_ NB FFP before gain
4 2 2 . '
| | ] =
4 a
4 = 3 a . .
NB FFP after gain
~d °
| ) 5 8 e o (Oo o 0 8 4 )
| T | r i ‘ T ' if ‘ i ‘ T

306 307 308 309 310 311 312 313 314

Frequency (MHz)

3.10: Noise power relative to the SQL as a

function of spectrum analyzer frequency. 50%
output coupling, 300 kHz resolution bandwidth.

30
a)
S 95
—J
*/
= 20
= 15
is
© 10
Q.

B 5
(@)

poe a tt ttt ttt fl tl

j r i n j L | rn J
NB FFP before gain |
a 6«66l --
a -a
P a a &
* _—
NB FFP after gain
ie) =
° 5 ° © 0 4 9o i
° L
i ; q ‘ q 7 l , q

306 307 308 309 310 311 312 313 314

Frequency (MHz)

Figure 3.11: Noise power relative to the SQL as a
function of spectrum analyzer frequency. 90% output
coupling, 300 kHz resolution bandwidth.

ses

28 ! i i { i 1 {oa

I 4 [

OG 1 C

” J

£ 18 4 -

£ 4

@ “J

2 137 -

@ J NB FFP before gain [

_ 7 ° r

gf 8 o ° C

o : ° Poy ° ° fe) L

® 3 _ o ° ° 0 ° Po

5 1 NB FFP after gain [

1 + + = ." i__4—__y
-2.. T T ;
306 307 308 309 310 311 312 313 314
Frequency (MHz)

Figure 3.12: Noise power relative to the SQL as a
function of spectrum analyzer frequency. 10% output
coupling, 300 kHz resolution bandwidth.

required to oscillate. The transmission of the 10% output coupler is 6.99 dB less than the
50% output coupler. Comparison of figures 3.10 and 3.12, for the case of the NB FFP
filter before the gain, show the relative noise power decreased by around 12 dB. Again,
the decreased EDFA gain required to oscillate resulted in lower amplified spontaneous
emission. For the case of the NB FFP filter after the gain and 10% output coupling, the
noise power relative to the SQL is within the experimental uncertainty of the SQL. The
laser output power with the intensity noise at the SQL was 200 mW.

In this chapter the intensity noise spectra of the fiber laser has been studied in detail.
At frequencies less than one-half of the free spectral range of the resonator, the dominant
intensity noise source is relaxation oscillation noise. This noise had a peak value of -105
dBc/Hz at the relaxation oscillation frequency which varies from 10 kHz to 35 kHz as the
output power varies from 0.1 mW to 1.0 mW. This noise rolls off at 20 dB/decade away
from this peak. At around 2 MHz the 20 dB/decade roll-off in the relaxation oscillation
noise begins to experience a noise floor due to beating noise between the lasing mode and
the strongly suppressed sidemodes.

At full power all sidemodes are suppressed by at least 60 dB. The sidemode
suppression increases rapidly for modes spaced farther from the lasing mode. At twice the
NB FFP bandwidth, the sidemodes have lost their strongly peaked nature and while still
visible are essentially broadband noise again. If the NB FFP is not placed before the
EDFA and the 50% output coupler, this broadband noise is around 20 dB above the shot
noise floor as measured with a balanced homodyne. If the NB FFP is between the EDFA
and the output coupler, the amplified spontaneous emission power can be reduced by 15
dB, relative to the SQL at a cost of a 3 dB reduction in laser output power, bringing the
measured noise to within 5 dB of the SQL for 50% output coupling.

The intensity noise due to beat noise can be further reduced by lowering the output
transmission coefficient. Lowering the output coupling decreases the intensity noise power

relative to the SQL by the same amount. An additional decrease is obtained due to the

decrease in the amplified spontaneous emission noise when the gain is decreased due to the
lower output coupling. However, the precise dependence of the noise power decrease due
to the gain decrease is complicated. For an output coupling of 10% and the NB FFP
between the EDFA and the output coupler, the intensity noise was reduced to the SQL for

frequencies greater than twice the NB FFP bandwidth.

References

{1] Amnon Yariv, /ntroduction to Optical Electronics, Third Ed. (Holt, Rinehart and

Winston, New York, 1985) p. 188

[2] Kerry J. Vahala, Dynamic and Spectral Features of Semiconductor Lasers, Ph.D. ’
dissertation, California Instit. Technol., Pasadena, CA (1985) |

[3] D. Stoler, Phys. Rev. D 1, 3217 (1970)

[4] L. Wu., H. J. Kimble, J. L. Hall, H. Wu, Phys. Rev. Lett. 57, 2520 (1986)

[5] R. E. Slusher, L. W. Hollberg, B. Yurke, J. C. Mertz, J. F. Valley, Phys. Rev. Lett.
55, 2409 (1985)

[6] G. J. Milburn, M. D. Levenson, R. M. Shelby, S. H. Perlmutter, R. G. DeVoe, D. F.
Walls, J. Opt. Soc. Am. B 4, 1476 (1987)

{7] B. L. Schumaker, Optics Letters 9, 189 (1984)

[8] E. Desurvire, IEEE Photon. Technol. Lett. 2, 208 (1990)

Chapter 4

Linewidth and Frequency Jitter Measurements
Using an Improved Delayed Self-Heterodyne

Interferometer

This chapter investigates the linewidth and frequency jitter of the single-frequency
fiber laser discussed in this thesis. A loss-compensated recirculating delayed self-
heterodyne interferometer (RDSHI) was conceived and demonstrated to permit accurate
measurement of the narrow laser linewidth with reasonable amounts of fiber delay. The
RDSHI will be described in detail. The linewidth of the fiber laser is then investigated
using the RDSHI. One of the advantages of the RDSHI is that it allows one to measure the
laser linewidth as a function of delay time. Based on this feature the fiber laser is shown to
have a linewidth dominated by frequency jitter. The timescale of this frequency jitter was

determined to be on the order of a millisecond.

4.1 An Improved Delayed Self-Heterodyne Interferometer for

Linewidth Measurements

The classic method for determining the linewidth of a laser is to construct two
lasers, stabilize them to an external reference cavity to remove long term drifts in relative
frequency, tune them close together and observe their inter-frequency beat note with a fast
photo-diode. This technique has two major drawbacks; the need for two independent,
tunable lasers and the need for frequency stabilization systems for both lasers. The delayed
self-heterodyne interferometer (DSHI) has been an important tool for the measurement of
laser linewidths since its conception [1], because it permits measurement of a laser's
linewidth without the drawbacks of the classic technique.

The conventional experimental set-up for a DSHI linewidth measurement is shown
in figure 4.1. The output of a laser is split by an acousto-optic modulator into a component

that has been frequency shifted by @, and a component that is essentially unaffected by

passage through the AOM except for attenuation. The frequency shifted component of the

light is coupled into an optical fiber and undergoes a large time delay Tp and is then sent to

a beamsplitter. At the beamsplitter, the time delayed, frequency shifted light recombines
with the unaffected beam. The two signals form a beatnote at @, on the photodiode. The
beat note can then be observed using an RF spectrum analyzer.

The basic idea in this experiment is that if the delay time of the long optical fiber is
longer than the coherence time of the laser, which is inversely proportional to the laser's
linewidth, then the time delayed signal will no longer have any phase relation to the
undelayed signal. Thus it acts as an independent laser source and the two signals will not
interfere coherently. With no relation between the phases, the beat note measured on the
spectrum analyzer is twice the laser linewidth in the case of a Lorentzian linewidth and

1.414 times the linewidth in the case of a Gaussian linewidth. If there were no delay and

Laser

AOM

Spectrum
Analyzer

Figure 4.1: The conventional delayed self-heterodyne
interferometer. AOM: acousto-optic modulator, + Ps
fiber delay line, PD: photodiode.

ee a

the signals were coherent, information conceming the laser linewidth would be difficult or
impossible to determine. The AOM serves mainly to move the beat note away from the
large noise floor inherent in any spectrum analyzer near DC. (Side note: This system uses
long lengths to measure narrow spectral profiles. At the other extreme, it is possible to
very accurately (10 um) measure distances using a source with a broad spectral profile.
This is known as optical coherence domain reflectometry [2].)

As the resolution of the DSHI is inversely proportional to the length of the delay
line, the DSHI trades off the drawback of two independent, frequency stabilized lasers for
a very long piece of fiber. For state-of-the-art semiconductor lasers, which typically have
linewidths on the order of hundreds of kHz to several MHz, this trade off is very
reasonable as a 5 km fiber delay yields a DHSI resolution of 40 kHz. However, for fiber
lasers, which have much narrower linewidths, the tradeoff is not so reasonable. In a recent
experiment a 72 km fiber delay was used to measure the linewidth of a fiber laser to be less
than 1.4 kHz [3] and the measurement was still limited by the length of the delay line. A
72 km, 1.5 pm fiber delay line at current market prices would cost $12,600 and have a net
loss of 14.4 dB. Increasing the resolution of the DSHI, another order of magnitude would
cost $126,000 and the delay line would have a net loss of 144 dB. Clearly, the trade-off
with the classic technique is beginning to favor two lasers.

Tsuchida [4] reported on an improvement to the DSHI method which uses a
recirculating delay, allowing the same fiber delay to be used multiple times. Placement of
the acousto-optic modulator in the delay arm of the recirculating DSHI (RDSHI) permitted
multiple delays to be determined by counting frequency shifts. However, due to large
losses, Tsuchida was only able to measure up to three passes through the fiber delay.

A significant improvement to the RDSHI may be obtained by including in the delay
arm an erbium-doped fiber amplifier. By partially compensating the large loss of the delay
arm with gain from the fiber amplifier, beat notes from light that has passed through the

delay as many as 30 times are easily discerned. An 11 km fiber delay line yields a

resolution limit of 18.2 kHz with a conventional DSHI. The loss-compensated RDSHI
yielded a resolution limit of 606 Hz for the same fiber length. If the 72 km delay line used
in the experiment mentioned above were inserted into this improved system, a resolution of
92 Hz would be obtainable.

The experimental setup for the loss-compensated RDSHI is shown in figure 4.2.
Components include an acousto-optic modulator (AOM), which provided a frequency shift
of 140 MHz with a conversion efficiency of 10% at 1550 nm; two fiber input/output
couplers to collimate the light out of the fiber for transmission through the AOM and then to
refocus it back into the fiber at the AOM output; a delay line consisting of an 11 km length
of optical fiber having a net loss of 0.2 dB/km in the wavelength range of interest; and a
fused fiber-optic coupler with a 90/10 coupling ratio. 90% of the light per pass was
retumed to the recirculator and 10% was sent to the photodiode. It is estimated that the
total loss per pass through the recirculator was 18 dB. Light was detected with an Ortel
photodiode (model 2515B) having a frequency response up to 15 GHz. This created the
potential to see light that had been delayed by as many as 100 passes through the fiber
delay line. The output of the photodiode was observed on a Tektronics spectrum analyzer
(model 2782) with a maximum bandwidth of 33 GHz.

In addition to these components which are standard in a conventional RDSHI, an
erbium-doped fiber amplifier (G) with two fiber-optic isolators having a reverse isolation of
35 dB each was added to the recirculator. Two commercially available amplifier units were
tested. One was a germanium-only codoped gain module capable of providing 40 dB of
small-signal gain at 1537 nm and a maximum saturation output power of 8.43 dBm. It
could provide sufficient gain to compensate the loss in the RDSHI delay arm over a
bandwidth of 6 nm about 1537 nm. The other module was an aluminium-germanium
codoped gain module capable of providing 37.2 dB of small signal gain at 1532 nm anda

maximum saturation output power of 10.3 dBm. It could provide sufficient gain to

10 %

Laser

Spectrum
Analyzer

Figure 4.2: The loss-compensated recirculating delayed
self-heterodyne interferometer. EDFA: erbium doped
fiber amplifier, AOM: acousto-optic modulator, T,:
fiber delay line, FFP (BB): fiber Fabry-Perot filter, PD:
photodiode.

compensate the loss in the RDSHI delay arm over a bandwidth of 30 nm from 1528 nm to
1558 nm.

The light beam diffracted by the AOM closes the recirculating loop. Thus with no
RF drive power to the AOM, the loop is open. With zero signal input to the RDSHI and
the AOM on, strong beat notes (>30 dB above the noise floor) spaced 18.2 kHz apart (the
free-spectral-range of the delay arm) were seen on the spectrum analyzer. This indicated
the recirculator was oscillating. This was obviously undesirable as it might interfere with
the measurement.

Two steps were taken to prevent the system from oscillating. First, a fiber Fabry-
Perot (FFP) filter was included in the delay arm. The FFP filter had a free-spectral-range
of 32 nm and a 3 dB bandwidth of 40.2 GHz. This inhibited broadband oscillation,
restricting it to the immediate spectral region surrounding the FFP filter transmission peak,
which was adjusted to transmit the signal wavelength. Second, the pump power to the
amplifier was adjusted downward until the signal input power was sufficient to saturate the
gain to a point below the oscillation threshold. This prevented oscillation while still greatly
reducing the net system loss. Taking these steps, spectrum analyzer beat notes now
occurred only at the expected 140 MHz spacing, with none observable at the 18.2 kHz
spacing indicative of oscillation.

It was observed that loud acoustic noises were capable of broadening the measured
linewidths (by a factor of about 2-5). To test whether this acoustic broadening was from
the measurement system or from the test laser, the entire 11 km fiber delay line was placed
in a styrofoam box to isolate it from these acoustic noises. There was no observed
decrease in the laser linewidth due to this system change. However, placing the test laser
in a similar box did produce a narrowing of the linewidth. Furthermore, after 22 passes
through the recirculator, the linewidth was observed to saturate at 4 kHz and did not keep
increasing as would be expected if the RDSHI was the cause of the broadening. It was

concluded that the measured broadening was due to the test laser.

In reference 4 the case of a RDSHI in which the net system loss is greater than 6 dB
is considered theoretically. This is clearly not the case for the loss-compensated system.
For high recirculator loss the main contribution to the power of the kth-order beat note is
from the undelayed signal field beating with the signal field that has passed through the
recirculator delay k-times. For the case of a loss-compensated recirculating delay it is
necessary to consider multiple contributions to the kth-order beat note. (For the purposes
of discussion, the case of the laser coherence time less than k times the recirculator round-
trip time will be considered.) In particular, the signal field that has been delayed byn+k
passes (n = 0, 1, 2...) can beat with a signal field that has been delayed by n passes to
generate a contribution to the kth-order beat-note intensity. It is therefore necessary to sum
over n, which in this case can be as large as 100. Provided that ktp is greater than the
laser coherence time, each of these contributions will have the same lineshape. The kth
beat-note under these conditions will therefore have a lineshape equivalent to a normal
DSHI with the same equivalent delay time.

Figures 4.3 and 4.4 show the photocurrent power spectra for various conditions.
In figure 4.3 the amplifier has been removed from the recirculator. Only a few AOM beat-
note peaks are seen. In figure 4.4 the amplifier has been inserted and many more AOM
beat-note peaks are visible. It is estimated that loss per pass has decreased from 18 dB in
the uncompensated case to approximately 0.2 dB per pass with compensation.

Figure 4.5 shows the measured FWHM linewidth of the laser as a function of
order. The RDSHI resolution as a function of order is also plotted. The spectrum analyzer
was used to average the linewidth over several seconds. The measured linewidth is
observed to increase with increasing order and ultimately saturate at 4 kHz around the 22nd
order. This behavior is characteristic of a laser with both a short term and long term
frequency stability and is discussed below.

However, before this increase could be attributed to the fiber laser, it was

necessary to eliminate the possibility of spectral broadening due to the erbium amplifier.

| Tek
Ref Lvl -10.0d8m _1008/ Atten COdB

EE ~~~ ee — ee —

Freq SOOM Span 1.0GHz
Res BW 300k Hz VidBW 300kHz SWP 63mS

LEVEL] [SPAN] Freq 50oMHz

Figure 4.3: Photocurrent power spectra of the
recirculator output in the case of no EDFA in the
recirculator. Only one beat note is clearly visible,
although three beat notes are present.

Figure 4.4: Photocurrent power spectra of the
recirculator output in the case of an EDFA in the
recirculator. Many beatnotes are visible.

Delay Time ( ms )

0 0.44 0.88 1.32 1.76
0 a

DSHI resolution

Detected Linewidth ( kHz )

2 E taal? |
L wi ee : }
0 aa | | | BOGGSSEEEEEE EE: 4
0 4 8 12 16 20 24 28 32 :
Order a

Figure 4.5: Measured linewidth of the fiber laser as a

function of order (lower axis). The RDSHI resolution
as a function of order is also plotted. Total delay time
of the signal may be read from the upper axis.

There is apparent disagreement in the literature [3, 5] as to whether erbium-doped fiber
amplifiers broaden the linewidth of a coherent source. To investigate further we removed
the 11 km delay line from the RDSHI in order to make k tp much less than the laser
coherence time [5]. The results are shown in figure 4.6. Figure 4.6 shows, from top to
bottom, the linewidth at 1st, 20th and 30th orders, respectively. The maximum observed
FWHM linewidth is less than 400 Hz indicating that spectral broadening from the amplifier
should not interfere with our measurement (the resolution limit of which is 606 Hz at 30
orders with the 11 km delay line).

As a further check, figure 4.7 shows the measured linewidth in the case of the
amplifier in the system (left column) and not in the system (right column). Only the first
three orders could be observed in both systems (1st order in the upper row, 3rd order in the
lowest row). The only discernable difference in the system with the amplifier and the
system without the amplifier was a decrease in the beatnote power due to the large
recirculator loss in the case with no amplifier.

The work described above was published in reference 6. A few months later Ishida
considered a nearly identical system [7]. Ishida used two acousto-optic modulators in his
system. One in the delay loop shifting the frequency by -200 MHz as above (except also
used to couple light out of the recirculator, as opposed to using the 90% output coupler),
and one before the recirculator shifting the frequency by -150 MHz. A 27 km dispersion
shifted fiber delay line was used in the recirculator. Use of two AOMs allowed separation
of the k-th beat note arising from the undelayed signal beating with the signal delayed k
times occurring at (200 k + 150) MHz from the component of the beatnote due to the sum
over n of the (n + k)-th delayed signal beating with the k-th delayed signal occurring at 200
k MHz. The former beat note was 15-20 dB bigger than the latter beat note. A third beat
note occurring at (200 (k + 1) - 150) MHz was attributed by Ishida to beating between the

undelayed, unshifted signal and signals due to four-wave mixing in the 27 km dispersion

lereeey

Bic pl
Ref Lvt -46.0d 10dB/ Atten OdB

Freq 140.000 715MHz~ Span 1.0KH:
ResBW 30Hz VidBw 30Hz SwP 6.35

{LEvEc] [SPAN] ResBw 3eHz

Ref Lvl -46.8dBm 10987 Atten OJB

Freq 2.800 014 2485Hz Spen 1.0hH: *
ResBwW 30Hz VidBwW 3OH7 SWP 6.35

{LeveL} [SPAN7} Span 1.@KHz

10dB/ Atten OdB

Freq 4.200 021 399GHz ~
ResBW 3OHz VidBW 30Hz
LEveC] [SPAN] ResBw 30nz

Figure 4.6: RDSHI output without the delay line. Top:
Ist order; middle: 20th order; bottom: 30th order.

Rel Lvl -34.0dBm 19d8/ Atten OdB

Freq 140.000 OOMHz Span SOH?
PesBW lkHz VidBW lkHz SWP 40005

(LEvEL] [SPAN] Ref Lvl -34.00Bm

Ref Lvl ~34_OdBm 103B/ Atten Odk

Freq 280.000 OOMH: : an SOK:
ResBW 1kHz VidBW IkHz SWPP 400mS

htver} [SPAN] Freq 280.000 core:

. Rind
Rel Lvl -34. OdBm 10dB/ Atten OdB

Spen SObH:
SwP 400mS

Freq 420.000 GOMHz —
ResBW kHz VidBW lkHz
Lever] [SPAN] Freq 420.000 earn:

Lvt -51.0dBm 10dB/_Atten Od _

wore

149 O00 OOMH,; Span FH.
WOE HY VidHw ikHe GW ache.

LtveL] [RESGw] Freq 140.000 B0MH-

Atton Odie

Span SO
SWE dikio.
LEVEL] fPESHW)] Freq 280.000 00MH,

sh
Bes ty +72. Odm 10387 Atten Gat:

ah - Awl

Freq 4202 000 @OMH, Span SOH.
PesRW liHz VidBW 1kHy SWPP adi’)

LEVEL] [RESBwW)] Freq 420.000 GoM,

Figure 4.7: RDSHI output with the EDFA (left) and
without the EDFA (right). Top row: 1st order, middle
row: 2nd order; bottom row: 3rd order.

shifted fiber [], which generated new frequencies at (50 +/- 200 k) MHz. The four-wave
mixing signals were 25-30 dB below the undelayed signal beating with the k-th beat note.
In the RDSHI used here to measure the fiber laser linewidth, the three beat notes
seen by Ishida are degenerate in frequency. However, those not due to four-wave mixing
have already been accounted for and would not generate an error in the linewidth
measurement. Furthermore, Ishida's fiber delay was 2.45 times as long as the delay used
here and the delay line was made from dispersion shifted fiber which has better four-wave
phase matching characteristics than the fiber used here. Even with the greater likelihood of
four-wave mixing in Ishida's experiment the four-wave mixing signal power was 30 dB
below the measurement signal power. Thus in the present experiment, four-wave mixing

should not introduce significant error in the measurement.

4.2 Linewidth and Frequency Jitter Measurement of a Single-

Frequency Fiber Laser

The linewidth of the fiber laser was measured between the wavelengths of 1531 nm
and 1538 nm, using the tenth order of the loss compensated RDSHI (corresponding to a
resolution of 1.2 kHz). The beat spectrum (figure 4.8) was taken in the averaging mode of
the spectrum analyzer over a period of several seconds with the resolution bandwidth
setting at 1 kHz. The full width at half maximum was calculated from the 3 dB and 6 dB
points of these spectra assuming a Lorentzian lineshape. An upper bound on the natural
linewidth was 2 kHz as given by the intersection of interferometer resolution and linewidth
data plotted in figure 4.5. There was no noticeable dependence on the laser wavelength
(Figure 4.9), laser power, or co-dopant of the erbium fiber used in the EDFRL. This
suggests that an extrinsic noise source is dominant over the linewidth of this laser. (The
Schawlow-Townes linewidth, for example, has an inverse dependence on the laser power

and gives a sub-Hertz value for the EDFRL at these power levels). Frequency jitter of the

. . ek
Ref Lvl -74.0dBm 5dB/ Atten OdB

— ——— —

: t
° a

Freq 1.400 @06 7O0GHz
ResBwW AkHz— VidBW 1lkHz SWP 400mS

(EVEL) [SPAN] SdBZ ts

Figure 4.8: Typical spectrum of the RDSHI at the 10th
order.

Laser Wavelength ( nm )

Figure 4.9: Measured linewidth of the fiber laser
between the wavelengths of 1531 and 1538 nm. Circles,
aluminum-codoped EDFA in the fiber laser; squares,
germanium-codoped EDFA in the fiber laser,

Sosa
ronan oes

detected beatnote induced mainly by acoustic or thermal cavity length fluctuations of the
fiber laser is the most likely source.

A laser linewidth that is broadened by frequency jitter due to cavity length
fluctuations may not obey the simple rule that the coherence time is inversely proportional
to the laser linewidth. To see this more clearly, consider figure 4.10, where laser
frequency is plotted as a function of time (in units of loop delay). In this figure the
thickness of the line representing the frequency is equivalent to the laser linewidth one
would expect from the Schawlow-Townes formula. If the dominant component of the
laser linewidth is frequency jitter, a single delay will be enough to see only the frequency
deviation corresponding to the average frequency jitter in one delay time. As this may be
bigger than the reciprocal of the delay time, one may erroneously conclude the full rms
laser linewidth has been measured. However, a truly independent laser has not been
simulated until the frequency jitter timescale has been exceeded by the fiber delay.

Therefore, in order to ensure that the frequency jitter component of the laser
linewidth can be measured, a delay line long enough to give time for the laser to exhibit its
full root mean square (rms) frequency variation must be placed in the conventional DSHI.
If the delay line is not long enough, then only a fraction of the full rms jitter will be
measured. Using the loss compensated RDSHI, the beat spectrum can be measured at
every order (up to 30th, see figure 4.5) to determine the dependence of the lineshape and
linewidth on the delay line length. The FWHM was again measured and calculated from the
3 dB and 6 GB point of the beat spectrum, assuming a Lorentzian lineshape. At higher
order a smooth transition from a Lorentzian lineshape to a Gaussian lineshape was
observed, as expected for the case of a lineshape dominated by frequency jitter. So the
linewidth at higher order was calculated based on the Gaussian lineshape. The laser
linewidth saturated to a constant value of approximately 4 kHz after 22nd order, indicating

that the frequency jitter timescale has been exceeded at this delay length.

100

Frequency Jitter after 1 delay time

Full RMS Frequency Jitter
(seen only after very long delay}

Frequency

Time (in units of delay) ;

SNe ante

Figure 4.10: One possible case of laser frequency vs.
time. Demonstrates RDSHI can be used to measure rms
frequency jitter on timescales of a few milliseconds.

101

For convenience the total delay time has been used to scale the top of the plot in
figure 4.5 and order number has been used to scale the bottom. From this plot it can be
seen that the frequency jitter timescale of the fiber laser is 1 ms and the rms frequency jitter
is 4 kHz. Furthermore, since the resolution of the measurement system to the natural
component of the linewidth is exceeded at around 2 kHz resolution, it can be safely

concluded that this component of the linewidth is less than 2 kHz. In order to measure this

component of the linewidth, the frequency jitter must be removed from the laser and thus
an active stabilization system must be employed. Active stabilization of the laser frequency

will be considered in the next chapter.

102

References

[1] T. Okoshi, K. Kikuchi, A. Nakayama, Electron. Lett. 16, 630 (1980)

[2] E. A. Swanson, D. Huang, M. R. Hee, J. G. Fujimoto, C. P. Lin, C. A. Puliafito,
Optics Letters 17, 151 (1992)

[3] H. Okamura, K. Iwatsuki, Electron. Lett. 26, 1965 (1990)
[4] H. Tsuchida, Optics Letters 15, 640 (1990)

[5] G. J. Cowle, P. R. Morkel, R. I. Laming, D. N. Payne, Electron. Lett.26, 424
(1990) .

[6] J. W. Dawson, N. K. Park, K. J. Vahala, IEEE Photon. Tech. Lett. 4, 1063 (1992)

[7] O. Ishida, IEEE Photon. Tech. Lett. 4, 1304 (1992)

103

Chapter 5

Frequency Stabilization of a Single-Frequency Fiber
Laser

This chapter discusses frequency locking of the single-frequency fiber laser to an
external fiber Fabry-Perot filter using the Pound-Drever technique [1]. As discussed in the
previous chapter, the linewidth of the fiber laser is dominated by frequency jitter. This
frequency jitter is believed to be due to cavity length fluctuations from thermal and acoustic
perturbations. These fluctuations cause the fiber laser linewidth to broaden to 4 kHz rms
on a timescale of 1 ms. On much longer timescales the laser frequency may be observed
(using the SuperCavity Fabry-Perot) to drift due to thermal length changes of the laser
cavity. Over a period of 10 sec to 1 minute, the laser frequency drifts about 2 MHz which
is sufficient to induce a mode hop.

It is desired to stabilize the laser frequency at least to within the 4 kHz measured
linewidth and to eliminate mode hopping due to thermal drift of the cavity. In this chapter,
the Pound-Drever technique will be reviewed, changes to the fiber laser cavity necessary to
implement this system will be presented, and, finally, possible improvements to the

stabilization technique will be considered.

104

5.1 Frequency Stabilization

Frequency stabilization of lasers has been a subject of active research since the early
1960's [2]. Many different methods have been proposed. Locking of the laser frequency
to an atomic absorption line is very popular as it allows absolute stabilization of the
frequency providing the temperature and pressure of the atomic cell is kept reasonably
constant [3, 4, 5]. While an atomic line provides an absolute reference, it may be desirable
to stabilize the laser frequency at a point that is some distance away from any convenient
atomic standard, especially in the case of a laser source that is continuously tunable over a
broad range.

Other stabilization methods, which provide greater flexibility in establishing
reference points, include locking the laser frequency to the side of an interferometer fringe
such as a Michelson or Fabry-Perot interferometer [6] or, alternatively, to the transmission
peak [1, 7]. To lock to the transmission peak of a Fabry-Perot filter, it is necessary to find
a way to determine which way the frequency is drifting. One way to do this is to dither
either the laser frequency or the Fabry-Perot transmission peak. Intentionally dithering the
parameter one is trying to stabilize is clearly not desirable in terms of obtaining a very stable
source. However, dithering the Fabry-Perot is often a reasonable way to proceed.

The Pound-Drever technique offers several advantages over simply dithering the
Fabry-Perot transmission peak. First, it is relatively simple to lock multiple lasers to the
same reference cavity using this technique [8]. Second, if a Fabry-Perot is dithered to
provide an error signal, the decay time of the optical cavity will limit the bandwidth of the
error signal. However, using the Pound-Drever technique, the Fabry-Perot acts as a
frequency discriminator on timescales longer than the cavity decay time and a phase

discriminator on timescales faster than the cavity decay time [1]. This makes it possible to

Sa eee

awa

105

phase lock two lasers to a Fabry-Perot cavity and achieve inter-frequency linewidths as
small as 3 Hz [9].

While feedback bandwidths to the fiber laser have not yet exceeded the external
cavity bandwidth, it was considered that this may be a desirable goal for future
experiments. Furthermore, locking multiple lasers to the same reference cavity is desirable
for use in wavelength division multiplexed fiber sensor systems, communications systems
or for high resolution spectroscopic applications.

The typical experimental set-up for obtaining a Pound-Drever error signal is shown
in figure 5.1. The output of a single-frequency laser is sent through an isolator to a phase
modulator. The phase modulator is driven by an RF oscillator at a frequency, Opm>
greater than two or three times the bandwidth of the Fabry-Perot being used. The laser
spectrum now consists of the lasing mode and two sidebands at +/- Wpyy. The phase
modulated light is then sent through a beamsplitter to the Fabry-Perot. (A polarization
selective beamsplitter in combination with a quarter wave plate or an optical circulator may
be used here for improved system performance.) Under ideal conditions, if the laser
frequency is tuned so that it coincides precisely with the transmission peak of the Fabry-
Perot, the lasing mode is fully transmitted. No part of the lasing mode is seen in the
reflected signal in this case. The phase modulated sidebands are outside the transmission
bandwidth of the Fabry-Perot and thus are fully reflected. The reflected signal is diverted
by the beamsplitter to a photodiode.

If the lasing mode is not tuned precisely to the transmission peak of the Fabry-
Perot, a small amount of the field will be reflected. Thus the signal arriving at the
photodiode consists of the phase modulated sidebands and a small portion of the lasing
mode. The lasing mode, in this case however, has undergone a phase shift with respect to
the phase sidebands due to being reflected from near the transmission peak of the Fabry-

Perot. A beat note at @pyy is formed between the modulated sidebands and the partially

reflected, phase shifted lasing mode at the photodiode. Phase sensitive detection of this

106

Laser

Fabry- | >
Perot Output

PID Feedback
Circuit

Figure 5.1: Pound-Drever technique for frequency
locking a laser to an external Fabry-Perot cavity.
PM: phase modulator; OSC: RF oscillator; MIX:
RF mixer; PD: photodiode,

107

“beat note via mixing with the original oscillator signal yields an error signal that is -
proportional in sign and amplitude to the deviation of the laser frequency from the
transmission peak of the Fabry-Perot.

If the laser phase changes much faster than the decay time of the Fabry-Perot
cavity, the signal reflected from the input mirror of the Fabry-Perot will not be in precise
anti-phase with the field stored in the cavity. Therefore, a portion of the field stored in the
cavity, which has a 7 phase shift with respect to the field reflected off the input to the
Fabry-Perot, will arrive at the photodiode and beat with the sidemodes forming an error
signal. This error signal is proportional in sign and amplitude to the laser's phase change.
Thus, the system acts as a frequency discriminator on timescales slow in comparison to the
cavity decay time and a phase discriminator on timescales faster than the cavity decay time.
The transition between these two regimes has been shown to be smooth [1]. Thus, a servo
system can operate properly with a response time far shorter than the cavity decay time
when allowance is made in the feedback electronics for a 1/2 phase change associated with
the transition [1].

The error signal generated by the Pound-Drever technique is typically processed by
proportional-integral-differential control electronics and fed back to the laser. Noise in the
error signal system will limit the accuracy to which frequency deviations can be measured.
In fact, noise in the error signal will be processed by the control electronics and impose
frequency fluctuations on the laser. For a system limited only by the shot noise at the

photodetector, the minimum achievable frequency jitter is given by [1],

dv(T) 2Sv, Paw (5.1)
Nt

108

where SV, is the bandwidth of the Fabry-Perot, Po is the laser input power, T the Fabry-

Perot transmission efficiency at the transmission peak, 1) is the quantum efficiency of the

detector, B=1/20T is the measurement bandwidth, and T>1/278V¢.

5.2 Application of the Pound-Drever Technique to a Single-

Frequency Fiber Laser

Consider the problem of frequency stabilization of the fiber laser. For the fiber
laser's frequency to be capable of remaining locked to an external Fabry-Perot, the
transmission peak of the internal NB FFP filter, which acts as the lasing mode selector,
must also track the external Fabry-Perot transmission peak. Figure 5.2 illustrates this
problem. |

To solve this problem, two error signals and control circuits are required, one for
the lasing mode and one for the internal mode selector. The Pound-Drever technique can
be applied in both cases. By placing the phase modulator in the laser cavity prior to the
output coupler and the NB FFP (see figure 5.3), an error signal may be obtained. This
error signal is proportional to the frequency difference between the lasing mode and the
transmission peak of the internal mode selector. Feedback to the internal mode selector
permits tracking of the lasing mode thus eliminating mode hopping completely. If the
lasing mode is then locked to an external cavity, the internal mode selector will be locked to
the lasing mode and thus also will be locked to the external cavity.

The lasing spectrum emitted from the laser in this case already contains the phase
modulated sidebands necessary for Pound-Drever locking of the lasing mode to an external
cavity. The only further components required are an external Fabry-Perot, a 50/50 fused
fiber coupler, another photodiode and a duplicate set of electronics. Controlling the lasing
mode requires control of the cavity length. In bulk solid state, dye or gas lasers this is

typically accomplished using a PZT. However, in the fiber laser, which has a much longer

109

External
NB FFP
Internal
NB FFP

™,

Lasing Mode

Nearest Side Modes

SS eee

Figure 5.2: An internal tracking filter is required to
achieve active locking to an external reference.

110

300 MHz

Mixer MCF
FFP
(MS)
D2
3 dB
D1
3 dB '
FFP
| (Ref)
FFP
(MS)
9 Mod OUTPUT
PC MCF
Gain (Er) (BB)

ISO

Figure 5.3: Experimental set-up for active stabilization
of the fiber laser frequency. FFP(MS): the internal NB
FFP(mode selector); D1, D2: photodetectors; ISO:

isolators; MCF: metal clad fiber; @ Mod: phase
modulator; PC: polarization controller; FFP (ref):
external NB FFP(teference cavity).

111

cavity length, a PZT does not provide sufficient excursion to compensate for thermal
expansion and contraction of the cavity due to room temperature changes of several
degrees. To provide sufficient excursion for the feedback process, changes to the cavity
length are accomplished via resistive heating of a metal clad fiber.

Again, figure 5.3 shows the experimental configuration used to implement the
Pound-Drever system in all fiber form and lock the lasing mode to an external reference.
The lithium niobate phase modulator placed prior to the output coupler was a United
Technologies model APE PM - 1.5 - 1.0 - XX - 01. It was pigtailed with polarization
preserving fiber, had a half-wave voltage of 6.6 V, and an insertion loss of 4.1 dB. It also
acted as the polarizing element in the laser. The NB FFP filter located inside the resonator
but after the output coupler acted as the mode selector for the laser. It had a free spectral
range of 5.92 GHz, a finesse of 133, a bandwidth of 45 MHz and was tunable across one
free spectral range by applying 16 VDC. The 3 dB output coupler acted as the beamsplitter
between these two devices. The reflected signal containing the difference frequency
information between the lasing mode and the NB FFP transmission peak was incident on
detector D1. This signal was amplified with a 50 dB RF amplifier, sent through a high
pass filter and mixed with part of the 300 MHz signal driving the phase modulator. BNC
cables of varying lengths were introduced between the high pass filter and the mixer in
order to achieve maximum phase matching between the signal and the local oscillator. A
proportional integral differential control circuit was constructed from op-amps to process
the error signal output from the mixer and feed it back to the internal NB FFP. This circuit
had a bandwidth of 10 kHz.

The output of the fiber laser was sent through an optical isolator to another 3 dB
fused fiber coupler. One output port of the coupler was sent to a fiber Fabry-Perot filter
closely matched in free spectral range and finesse to the internal mode selector. This was
done so that the laser could be tuned by the BB FFP filter and the new lasing frequency

could easily acquire a lock to the external reference. It was also tunable, allowing the

112

lasing frequency to be fine tuned via tuning the external reference. The other output port of
the 3 dB coupler was carefully terminated by tying a knot in it to produce large bending
loss and dipping the end in an index matching gel. This prevented stray reflections from
creating a Michelson interferometer effect and interfering with the error signal. The
reflected signal from the reference FFP was incident on photodiode D2. The control circuit
electronics were similar to those described above.

Again a proportional-differential-integral control circuit was constructed, only in
this case, the voltage output was connected to a voltage controlled current source. This
current source was capable of supplying 0.5 amps of current to the metal clad fiber. The
metal clad fiber inserted in the cavity was a standard single mode fiber only coated with
aluminium. It had a length of one meter, a resistance of about 2 Q and an insertion loss of
3 dB. With the metal clad fiber inserted in the cavity, 0.5 amps of current applied, and
with the NB FFP tracking circuit active, the lasing mode was observed to move about 200
MHz in one second. This indicated that the metal clad fiber could compensate for cavity
length excursions as large as 50 um and had a response time on the order of 5 ms/MHz.

Based on the frequency jitter timescale measured in the last chapter, this is sufficient
to stabilize the laser frequency to within its frequency jitter linewidth. From equation 5.1,
for operation at 1550 nm, a measurement bandwidth of 10 kHz, an optical power of 1 mW,
a transmission efficiency of 50%, a cavity bandwidth of 45 MHz and an overall quantum
efficiency in the system of approximately 20% due to losses in the system, the minimum
frequency fluctuation obtainable is 160 Hz. However, since the laser is not in the
configuration for shot noise limited operation, the actual noise current at the photodetector
may be as much as 15 GB larger giving a minimum theoretical frequency fluctuation that is
closer to 5 kHz. Possible ways to improve this limit will be discussed below.

The stabilization system was implemented. The error signal from mixer 1] is shown
in figure 5.4; the upper curve is for the control loop closed and the lower curve is for the

control loop open. Without feedback the error signal shows frequent modehops with

113

Frequency Deviation

Time ( minutes )

Figure 5.4: Error signal from the internal NB FFP. No
feedback (lower curve); with feedback (upper curve).

114

frequency deviations from the transmission peak of the mode selector as large as +/- 5
MHz. This deviation is somewhat larger than expected (+/- 2 MHz) and is attributed to
another weak Fabry Perot being formed between splices, as discussed in Chapter 3. With
feedback, however, modehops are completely eliminated. This was confirmed by the chart
recording shown in figure 5.4, observation of the lasing mode with the SuperCavity and
observation of the error signal on shorter timescales with an oscilloscope. The oscilloscope
provided resolution of the error signal to 1 mV, which corresponds to 100 kHz of
frequency deviation. In addition, perturbing the system by varying the current to the metal
clad fiber and banging on the optical table to induce strong acoustic perturbations confirmed
that this loop of the stabilization system was quite robust.

The error signal from the external fiber Fabry-Perot is shown in figure 5.5, the
upper curve is for the control loop closed and the lower curve is for the control loop open.
Without feedback to the metal clad fiber, but with the first control circuit operating, the
deviation between the lasing frequency and the transmission peak of the external Fabry-
Perot is approximately +/- 15 MHz over a time period of 10 minutes. With feedback, this
deviation is greatly reduced. Again, this was confirmed by observation of the lasing mode
with the SuperCavity and observation of the error signal on shorter timescales with an
oscilloscope. The oscilloscope resolution was 1 mV, which corresponded to 200 kHz of
frequency deviation from the transmission peak of the external Fabry-Perot (the optical
power is lower outside the cavity, thus the error signal is smaller). The external Fabry-
Perot was tuned and mode hop free tracking of the transmission peak was observed with
the SuperCavity over the 200 MHz range of the metal clad fiber. Again, banging on the
optical table to induce strong acoustic perturbations showed this loop of the stabilization
system was also quite robust. While it has been shown using the error signal that the
frequency fluctuations of the laser have been reduced below 200 kHz, actual measurement
of the stability must wait until components for two completely stabilized lasers may be

obtained. The experimental set-up for this experiment is shown in figure 5.6.

Frequency Deviation

Time ( minutes )

Figure 5.5: Error signal from the external refernce NB
FFP. No feedback to MCF, but with feedback to |
internal NB FFP (lower curve); with feedback to both
(upper curve).

116
300 MHz
Band Pass
Filter
300 MHz \
Detector
Laser 1 Optical

Circulator

Optical
Coupler '

Laser 2 FFP
eo] LL

OUTPUT

350 MHz 4
350 MHz
Band Pass
Filter

Figure 5.6: Experimental set-up to stabilize two fiber
lasers to the same reference FFP (FFP (Ref)).

117

This experiment has demonstrated the feasibility of active stabilization of the fiber
laser frequency. There are many improvements that could be made to future systems. For
example, shortening the cavity length by cutting the pigtails on the equipment and using a
shorter, more highly doped erbium fiber gain medium would increase the ease of achieving
stabilization and increase the continuous tuning range of the laser. While finite excursion
of the metal clad fiber will limit the continuous tuning range, it is anticipated that the quasi-
continuous tuning or seamless tuning with stitching will be possible. The metal clad fiber
used in this experiment changes the cavity length by 50 free spectral ranges. With a cavity
length 10 times shorter, the continuous tuning range would be 10 times as large.

Tracking of the lasing mode with the NB FFP could also be accomplished by
dithering the NB FFP to create an error signal. This would allow the phase modulator to
be removed from the cavity and the NB FFP to again be placed prior to the output coupler,
lowering the noise in the system to within 5 dB of the shot noise limit. Alternatively, or in
conjunction with that method, a higher finesse Fabry-Perot, such as the SuperCavity, could
be used to reduce the minimum frequency jitter by a large amount. It may be desirable to
leave the phase modulator in the cavity since its large bandwidth would be useful to employ
the phase discriminator aspects of the Pound-Drever technique.

From the results in this chapter and the previous three chapters, it is clear that the
fiber laser has the potential to be a very stable, low noise, widely tunable source in the 1.5

Lim wavelength region.

118

References

[1] R. W. P. Drever, J. L. Hall, F. V. Kowalski, J. Hough, G. M. Ford, A. J. Munley,

H. Ward, Appl. Phys. B 31, 97 (1983)

[2] A. D. White, JEEE J. Quantum Elec. QE-1, 349 (1965)

[3] S. L. Gilbert, Optics Lett. 16, 150 (1991)

[4] M. Suzuki, S. Yamaguchi, JEEE J. Quantum Elec. QE-24, 2392 (1988)

[5] M. Arditi, T. R. Carvet, Proc. IEEE 51, 190 (1963)

[6] E. A. Ballik, Phys. Lett. 4, 173 (1963)

[7] A. D. White, EEE J. Quantum Elec. QE-1, 322 (1965)

[8] B. Glance, et al., Electron. Lett. 23, 750 (1987)

119

[9] T. Day, E. K. Gustafson, R. L. Byer, Optics Lett. 15, 221 (1990)

120

Chapter 6

Multiple Wavelength Fiber Lasers

The previous five chapters have investigated single-frequency operation of an
erbium doped fiber laser. Broad tunability, narrow linewidth, low intensity noise and
active frequency stabilization have been demonstrated. Applications such as semiconductor
four-wave mixing spectroscopy that take advantage of these properties are beginning to
prove the laser's usefulness [1]. However, if the fiber laser were to be used in a
wavelength division multiplexed telecommunications or sensor application, the cost would
need to be greatly reduced in order to be competitive with commercial DFB lasers or even
external cavity semiconductor lasers. The two major barriers here are the price of the pump
diode and the price of the fiber Fabry-Perot and the optical isolators.

Commercial 980 nm pump diodes with output powers of 1 W are now available. A
fiber laser can achieve a 1 mW output power (sufficient for many uses) over a broad tuning
range with approximately 25 mW of pump power. This creates the possibility that 40 fiber
lasers could be pumped with a single pump laser. Most of the cost would then be
contained in components other than the gain media. So the question is how to reduce the
cost of the components. Alternatively one might ask how to reduce the number of

components. If the lasers can share the same location, they could conceivably share parts

Soo =

121

of the same cavity, creating, in effect, one laser with multiple single-frequency outputs.

This is the subject of this chapter.

Two means of creating multiple frequency lasers were investigated. In the first,
carried out prior to the rapid drop in price of the 980 nm pump laser, the lasers share the
same gain medium. Weak spectral hole burning in combination with careful matching of
the loss to the gain at each wavelength allow multiple lasers to share the same gain medium.
In the second, each laser has an independent gain medium and as many of the remaining
cavity components as possible are shared.

The first section of this chapter investigates spectral hole burning in EDFAs. For
most practical purposes the EDFA gain medium is homogeneously broadened and spectral
hole burning can be safely ignored. However, a small amount of weak spectral hole
burning is observable [2]. The second section investigates the case of the two-frequency
fiber laser. The two-frequency fiber laser is studied both for the case of a shared gain
medium and, two, independent gain media. The third section presents the results from a
demonstration of a six-frequency, single gain medium fiber laser. The results of the
second and third sections, in combination with the results of the previous four chapters are
used to propose an N frequency, low intensity noise, narrow linewidth, frequency

stabilized fiber laser based on state-of-the-art fiber optic technology.

6.1 Spectral Hole Burning in Erbium Doped Fiber Amplifiers

The gain spectrum of the erbium doped fiber amplifier has been extensively
investigated [2, 3, 4, 5, 6, 7]. The broad gain bandwidth of the EDFA can be mainly
attributed to the Stark splitting of the upper and lower levels of the 1550 nm transition (see
figure 6.1), which creates a large number of transitions from 1470 nm to 1570 nm. The
Stark splitting of the upper and lower energy levels results in a manifold of energy states at

each level, which are populated in a Boltzmann distribution [5]. The Boltzmann

122

6800F 6770 (222-230)
= 6700K 14 #6711 (163-171)
- 6600 i j 6644 (96-104)
o - 6540-48 (0-8)
= oor PITT FIT 4
> aX ABER YESS BASSE
c T HMHTT NMHNY nn nn 3
ul S00 aT AE OAT AT DT DB aren Gees Dee Oe 268
200K 201
= @-@-6-@ —_—_—— |-
oF -@006-06 —¥O
(26-34)
6800F 6770 (222-230)
| 6600F
= O88 eC TETP errr
- ws mNNNNNN TN NMNM 3
G6 AW BWA BBe ee aes He
200K a Y u Y 201
100k v v v 125-133
5 E ~y +~+—_— S| 59

(26-34)

Figure 6.1: Energy level diagram showing Stark
components of the upper and lower energy levels of the
EDFA. Observed fluorescence transitions (lower);
observed absorption transitions (upper). Source [5].

123

distribution of the energy levels results in absorption in the 1470-1520 nm range and
fluorescence/gain in the 1520-1570 nm range when the amplifier is strongly pumped.

One might expect that due to the Stark split energy levels, strong spectral hole
burning would be observed even at room temperature. However, the resulting energy
levels are very closely spaced in energy (~50-100 cm!) and interaction with phonons in
the glass causes a fast cross-relaxation between levels (which also results in a broad
homogeneous linewidth for each level). As a result of this fast cross-relaxation, strong
spectral holes are not observed at room temperature [3, 6]. However, at lower
temperatures, the number and average energy of the phonons in the glass is greatly
decreased and it becomes possible to observe spectral hole burning in EDFAs [3, 6]. As
the temperature is increased from 7 K to 61 K, the observed spectral hole width increases
from 0.21 nm to 1.3 nm. At room temperature the homogeneous linewidth of any
transition between the upper and lower manifold is extrapolated to be on the order of 23
nm.

The cross relaxation process keeps the Boltzmann distribution of the electron
populations in the Stark levels in rough equilibrium in the presence of strong saturating
signals. As a result, gain cross saturation is observed; the gain at short wavelengths
saturates more easily than the gain at longer wavelengths [2]. However, due to the finite
rate of these processes, weak spectral holes at the saturating signal wavelengths are
observed in the gain compression spectra [2]. These holes have a depth of about 1 dB and
a width of about 1 nm.

To verify this result and investigate hole burning further, two single frequency fiber
lasers were constructed and used to measure the small signal gain of the EDFA in the
presence of a saturating signal of varying power and wavelength. The experimental set-up
is shown in figure 6.2. Fiber laser #1 creates the saturating signal which is sent through a
50/50 fused fiber coupler to a Corning FiberGain module (EDFA), which is a germanium

codoped amplifier with 37.1 dB of small signal gain at the gain peak (1537 nm) and a

124

i #1
Fiber Laser 50/50

Chopper

Fiber Laser #2

ATT

a Current
Lock-in | Meter }a
PD

EDFA

Lock-in Current
Meter

SuperCavity

Spectrometer

Figure 6.2: Experimental set-up for measuring gain cross-
saturation and spectral hole burning. ATT: variable
attenuator; PD: photo-diode.

125

saturation output power of +8.3 dBm. Fiber laser #2 acts as the probe to measure the small
signal gain of the EDFA as a function of wavelength. The output of fiber laser #2 is
collimated by a fiber input output coupler, attenuated to small signal power levels (~-35 to -
40 dBm) and chopped to allow lock-in detection to distinguish the probe signal from the
much stronger pump signal. The chopped, attenuated light is recoupled back into fiber by a
second fiber input/output coupler and sent through the 50/50 coupler. One half goes to the
EDFA and the other half is detected with a broad area Ge photodetector driving a lock-in
amplifier.

The output signal from the EDFA is split by a 50/50 coupler. One-half goes to
another Ge photodetector and a second lock-in measures the output probe power. The
other half is sent to a 90/10 fused fiber coupler, 10% of this power is used to monitor
stability of the two fiber lasers and 90% is sent to a spectrometer to monitor the
wavelength. The signals from the two lock-in amplifiers are divided by an analog voltage
divider and the gain (less a several dB correction factor accounting for coupling losses in
the system) may be read directly from the output of this divider.

Figure 6.3 plots EDFA gain as a function of wavelength for the cases of no applied
saturating signal, a -20 dBm saturating signal applied at 1536.6 nm and a -20 dBm
saturating signal applied at 1537 nm. Gain cross saturation can be clearly observed. With
the 1536.6 nm signal applied, the gain at 1538.6 is nearly unaffected, but with the signal
applied at 1537, the gain at 1538.6 is down by 3 dB. At shorter wavelengths the gain is
quite close to the same value in both cases. The amount of gain compression at shorter
wavelengths is apparently decreasing as the probe is moved further from the saturating
signal wavelength. Gain compression is ~5 dB at the saturating signal wavelength and
decreased to around ~3 dB one nanometer to the blue side of the saturating signal.

Figure 6.4 shows the small signal gain spectrum as a function of saturating signal
power applied at 1537.2 nm. As the saturating signal power is increased, the gain peak is

steadily compressed. It can be seen that while the gain of the EDFA is essentially

126

40 [ T T ~T T T T T fF T t T T a
i: © No pump 4
35 | ee :
i wD 2 o 7
[ o 0 0 6 Z
30 + a x3 xX x On OX :
a V ° } “ |
ij o © >» 2 * x OO 69 fons 1
a r wx | 4
oS C 4
a i: © Pump (0.011 mW) =X Pump (0.011 mW) 4
= 20
ig) L |
‘O) ' 4
15 :
io. fF: |
5 |
0 i i k i i Ll 1 1 Lt L i 1 3 1 i i i
1535 1536 1537 1538 1538 ;
Wavelength (nm)

Figure 6.3: EDFA small signal gain vs. wavelength in the
case of no other signal (circles); -20 dBm signal at 1536.6
nm (diamonds); -20 dBm signal at 1537 nm (X).

127

40 :

35 - :

30 | ch Reed Ga :

F .° e J

L e e 4

ay 25 :

c 20 oOo ~ ° a : : nn -— - " oa

se) c :

10 | ee, ° , KK ge ne

L X x x x x A

5 i

0 :
1535 1536 1537 1538 1539

Wavelength (nm)

Spraiapcee earaa

Figure 6.4: EDFA small signal gain vs. wavelength in
the case of no other signal (circles); signal at 1537 nm
at -11 dBm (squares); -5.3 dBm (diamonds);

-1.1 dBm (X)

128

homogeneously broadened in that a saturating signal at one wavelength will saturate the
amplifier across the entire gain spectrum, this saturation is not completely uniform.
Multiple-wavelength operation of a laser would be difficult in an ideal homogeneously
broadened gain medium. However, it is clear from the observation of gain cross-saturation

that the EDFA is not an ideal homogeneously broadened gain medium.

6.2 Co-Lasing in an Electronically Tunable Fiber Laser

A dual-frequency, co-lasing, widely-tunable laser source has constructed from
essentially the same components as the single-frequency source [8]. By using a single
ring, however, a reduction in the number of components that would be needed to construct
two separate, single-frequency sources is achieved.

Two experimental configurations for obtaining co-lasing operation were
investigated. They are shown in figures 6.5a and 6.5b. In figure 6.5a we see the single
gain module configuration. The gain module (G) is a commercial: erbium doped fiber
amplifier consisting of approximately 20 m of fiber. The index raising codopants included
aluminum. The erbium doped fiber amplifier was pumped with a 980 nm laser diode and
provided up to 37.2 dB of small signal gain and 10.3 dBm of maximum output power at
1532 nm, It was possible to achieve lasing over the entire region accessible with the tuning
filters.

The two tuning filters were placed in the arms of a Mach-Zehnder interferometer
created by two 3 dB fused fiber couplers. A calculation discussed below showed that the
minimum frequency separation between the co-lasing frequencies was limited by the
bandwidth of the tuning filters. The tuning filters were broadband Micron Optics Fiber
Fabry-Perot filters (BB FFP). One had a free spectral range (FSR) of 4020 GHz, a
bandwidth of 26.1 GHz, and an insertion loss of less than 3 dB. The other had a FSR of
4700 GHz, a bandwidth of 38.2 GHz and an insertion loss of less than 2 dB. It is

129

(a)

Figure 6.5: Dual-frequency fiber laser with one gain
medium (a) and two independent gain media (b). G:
EDFA; Pol: polarizer; PC: polarization controller; NB
FFP: narrowband fiber Fabry-Perot; BB FFP:
broadband FFP; ISO: isolator.

130

estimated that the total cavity length for either one of the two wavelengths was
approximately 40 m corresponding to a mode spacing of 5 MHz. The total minimum loss
seen by one wavelength was about 12 dB.

Co-lasing was achieved by tuning the BB FFP's to the desired wavelengths and
then using the polarization controllers in combination with the polarizer to balance the
losses with the amplifier gain. Output from the laser was split by a 3 dB coupler. Half the
output was sent to a spectrometer and the other half to a scanning Fabry-Perot
Interferometer (Newport Research Super-Cavity SR-170 FSR 6 GHz, resolution 1 MHz).

The Super-Cavity Fabry-Perot Interferometer showed that the configuration in
figure 6.5a operated with only one longitudinal mode excited at each wavelength (see figure
6.6). The smaller peaks in the picture are transverse modes of the Super Cavity Fabry-
Perot. Figure 6.7 shows a plot of the achievable tuning range for this configuration
measured using the grating spectrometer. Each point on the graph represents one measured
spectrum taken with FFP voltages fixed. Tuning was limited only by the free spectral
range of the BB FFP's. Some difficulty in tuning was encountered around 1537 nm. Itis
believed that this is due to a system idiosyncrasy currently under investigation. When the
filters were tuned into the same wavelength region and close enough in wavelength to
create significant bandpass overlap (proximity tuning), a Mach-Zehnder effect occurred
reducing apparent cavity loss to about 5 dB for a single mode. In this case the laser
reverted to single-frequency operation.

Maximum output power per wavelength for co-lasing operation was approximately
200 uW. For proximity-tuning induced single wavelength operation, the power into the
single lasing mode increased by more than a factor of two as compared to the power per
mode measured for co-lasing operation. This increase was the result of the obvious
increase (by an approximate factor of 2) in the quantum efficiency per mode and a decrease

in lasing threshold caused by the apparent reduction in Mach-Zehnder loss mentioned

above.

131

eee

Figure 6.6: Typical output of the Newport Supercavity
spectrum analyzer, showing two co-lasing modes
separated by several nanometers in wavelength but
folded over by the 6 GHz FSR of the Supercavity. The
smaller peaks are the transverse modes of the
Supercavity.

132

1570
1560- eee oe © £000 8086 8 eee
~ 4 ee = ° eo ef @ @ ee : eo eae
£ 1550 7 > © @ * e oeee# ie a ee
c ° oe
— ee eee e eeoeeee8 @ #eeese
Q , ° :
< 1540 oe. Ps eo ee e@ @ g@ 8 ee ee @
1530 - eee ee e ee @ a e@eeee
1520 +--+ -

1520 1530 1540 1550 1560 1570
41 (nm)

Figure 6.7: Experimental tuning data from
configuration in figure 6.5a. Co-lasing was observed at
each point on the graph.

133

Figure 6.5b shows the other configuration investigated. It contained two gain
modules: the first being the module used above and the second gain module similar to the
first, however, with a pure germanium index raising codopant (i.e., no aluminium). The
second gain module had a small signal gain of 40.0 dB and a maximum output power of
8.43 dBm at 1536 nm. Tuning for this device was limited by the bandwidth of the second
gain module. The configuration in figure 6.5b also contains a narrow band (NB) FFP
filter. This device has a free spectral range of 10.4 GHz, a bandwidth of 130 MHz and an
insertion loss of 4 dB. It also has some polarization dependence due to its long cavity
length and high finesse, making the apparent fiber induced birefringence significant.

One of the advantages of the dual amplifier configuration is that there is no power
sharing between modes. A further advantage of this configuration is that there is no need
to adjust the polarization controllers to balance the gain and loss. The gain modules see
only one narrow wavelength range and they dynamically adjust to balance the loss. With
the NB FFP in place, the lasing modes have stability similar to the single frequency device.
However, due to the narrower gain bandwidth associated with the germanium only
codoped vs. the aluminium codoped gain module, the tuning area is reduced (See figure
6.8). It is apparent that the smaller gain bandwidth of the germanium codoped amplifier
provided sufficient gain to achieve lasing only in the region 1533 nm to 1541 nm around
the 1537 nm gain peak and in the region 1551 nm to 1555 nm around the 1555 nm gain
peak. The tuning range could be improved by using two aluminium codoped gain modules.
Another aluminium codoped gain module was not available at the time of this experiment.
Lowering the cavity loss would also increase the tuning range.

Proximity tuning was again restricted by the finite bandwidth of the BB FFP. This
phenomena has been investigated theoretically by considering transmission through a
Mach-Zehnder interferometer with Fabry-Perot interferometers in its arms. Two cases are
considered: nonoverlapping FFP filter resonances and overlapping FFP filter resonances.

By using parameters characteristic of our fiber Fabry-Perot filters and assuming a 12 cm

134

1560

Oo | oebesscesbssas cc,

2. 1550-

3 &

o =

9 &

6 &

- <2 {540 7 ae eo eee sory ese
1530 a —

1520 1530 1540 1550 1560 1570
11 : Al-Codoped :
Arm (nm) }

Figure 6.8: Experimental tuning data from
configuration in figure 6.5b. Co-lasing was
observed at each point on the graph.

135

path length difference for the Mach-Zehnder (the path length difference for the experimental
case was not easy to determine with great accuracy but was close to this value), the
transmission spectra shown in figure 6.9a and 6.9b were generated. Figure 6.9a is an
example of non-proximity-tuning showing no serious deformation of the individual peaks.
Co-lasing could be achieved in this regime. Figure 6.9b is an example of proximity-tuning
and the resulting interference effects. Single mode operation would be favored in this case.
In these plots, unity is full transmission through the interferometer. Insertion loss due to

the BB FFP's was not included in this simple calculation.

6.3 N-Frequency Lasers

The dual-frequency laser has demonstrated the feasibility of multiwavelength laser
operation. While the dual-frequency laser provides some cost reduction from sharing
components between two lasers, true cost reduction will only be achievable in the case of a
large number of frequencies. However, to scale up the number of frequencies, an
approach, based on an FFP in each arm and an N-star coupler will quickly become
unworkable in terms of total cavity loss and the number of FFP's required. To solve the
loss problem and provide coarse frequency selection, wavelength division multiplexers
were investigated [9].

Wavelength division multiplexers are presently a subject of active research due to
the widespread interest in wavelength division multiplexing for telecommunications
systems [10]. Two grating type wavelength division multiplexers were obtained from JDS
Fitel at a cost of $6,000 each. These multiplexers had 8 channels, a channel spacing of 4.8
nm, a bandwidth of 0.8 nm and an insertion loss of 6 dB per channel. Multiplexers similar
to these with up to 32 channels spaced 1 nm apart have also been constructed [10]. It is
expected that better devices at lower cost will be obtainable as WDM communications

systems become closer to realization.

136

1.0
0.8 T

0.6 f
(a)

0.4 +

1588 1590 1592 1594
1.0:

0.8 T

0.6 {

(b)

Normalized Transmission
Through the Mach-Zehnder

0.4 } |

Pw Nine.

1588 1590 ~—«1592 1594

Wavelength (nm)

0.2}

Figure 6.9: Theoretical calculations of Mach-Zender
throughput. (a) Non-proximity tuning and (b) proximity
tuning.

137

Figure 6.10 shows the simplest laser cavity configuration for a multichannel laser.
The optical gain was provided by a Corning FiberGain module which has 20 meters of
Aluminium codoped erbium-doped fiber pumped by a temperature-controlled 980 nm
diode. With variable output couplers made from fiber loop mirrors (FLM) placed as shown
in the figure, simultaneous lasing action was possible on all 8 channels by careful balancing
of the output coupling loss and the gain for each channel. The output powers in this case
ranged from 10 to 200 pW. However, a large power fluctuation was observed (about 3 dB
variation) on a millisecond time scale when a single channel output was monitored using a
photodiode and an oscilloscope. It is believed that this fluctuation results from modal
instabilities due to spatial hole burning by the intensity standing-wave pattern of
longitudinal modes well within the grating passband of each filter.

To test this hypothesis, and to eliminate the problem, a multichannel, traveling-
wave, ring laser was constructed by employing a second WDM, wavelength matched to the
first (figure 6.11). In this case, the variable cavity loss was provided by a polarization
controller on each channel combined with an in-line polarizer (plasmon wave type,
extinction ratio 24 dB). Due to the large losses and lack of spatial hole burning, lasing was
simultaneously achieved on only 6 channels. The measured output power was about 50 to
150 uW for each channel] and the total output power was approximately 1 mW. For this
cavity configuration, the power fluctuation was greatly decreased, to the point where
fluctuations were difficult to measure using the oscilloscope.

The spectrum shown in figure 6.12 was taken from the multiplexed output port of
the laser, using a grating monochronometer. It clearly shows 6 channel operation from
1528 nm to 1557 nm, with channel separation of 4.8 nm. The output power for the highest
peak (1537.8 nm) was about 150 uW at the demultiplexed output port. It was possible to
balance the output powers of all channels at the demultiplexed output ports by adjusting the

losses on each channel. Since it was also possible to achieve oscillation on the missing

138

CX ¢ Z
AL+AZ+...4+A8 FLM

Sos

Figure 6.10: Eight-channel laser configuration based on
a linear cavity. (FLM: fiber loop mirror; PC:
polarization controller; WDM: wavelength division
multiplexer; G: gain module.)

139

AL+A2+...+A8 Al

; Pol
10% | coo 4,
W W
Ye
OOO _J
50%

Ne G ace’

Figure 6.11: Eight-channel laser configuration based on
a ring cavity. (WDM: wavelength division multiplexer;
G: gain module; ISO: isolator; Pol: polarizer.)

140

200 T T T ! T T T
160 F 4
120 F 7
80 - 4

40 + | 4

@) ———

Output Power ( uW )

1528 1534 1542 1550 1558 1562
Wavelength ( nm )

Figure 6.12: Spectrum from the multiplexed output port
of a six-channel ring laser (figure 6.12). The output
power for the highest peak at 1537.8 nm was about

150 LW at the demultiplexed output port.

14]

channels in the spectrum (1542.7 nm, 1561.8 nm) in the other configuration, it should be
possible to achieve 8 channel operation using independent gain media.

Based on the information learned in the dual frequency and multifrequency fiber
laser experiments combined with information from Chapters 3, 4 and 5, an improved N-
frequency fiber laser design is proposed in figure 6.13. This design offers actively
stablized, low intensity noise single-frequency operation at N frequencies. Each frequency
is locked to the NB FFP and the NB FFP is controlled by measuring the error signal
obtained from an atomic source in combination with one of the frequencies located close to
that source. The shared output coupler would provide a low intensity noise output for the
N-frequencies. In this configuration all of the expensive components are shared by N
lasers.

Each of the N lasers would ideally have its own highly doped, aluminum codoped
EDFA, 2 fused fiber WDM couplers for combining the signal and pump, a metal clad fiber
for independent length control of each cavity and the necessary electronics for the
stabilization circuit. Total laser cost per channel would then depend on how densely the

wavelengths could be packed in combination with how much the shared components cost.

cies

“Seg ae

142

To feedback
electronics

nit

ces)
oO
un
124)
Lo)

fee)
ye
Kg

wo
co
co
ut
wn
oO

DOPOD

oo
un
wn

[ WDM
\ ox Tap 9
Ne,
10% Tap
NO;
10% Tap
Ne
10% Tap
Ne;
10% Tap
NO;

Figure 6.13: N-frequency fiber laser (proposed). WDM:
wavelength division multiplexer; PD: photodiode; Er:
erbium doped fiber; MCF: metal clad fiber; PM: phase
modulator.

143

References

[1] J. Zhou, N. K. Park, J. W. Dawson, K. J. Vahala, et al., to be published Appl. Phys.

Lett.

{2] M. Tachibana, R. I. Laming, P. R. Morkel, D. N. Payne, Optics Lett. 16, 1499
(1991)

[3] E. Desurvire, J. L. Zyskind, J. R. Simpson, /EEE Photon. Tech. Lett. 2, 246 (1990)

[4] C. R. Giles, D. DiGiovanni, JEEE Photon. Tech. Lett. 2,797 (1990)

[5] E. Desurvire, J. R. Simpson, Optics Lett. 15, 547 (1990)

[6] J. L. Zyskind, E. Desurvire, J. W. Sulhoff, D. J. DiGiovanni, JEEE Photon. Tech.
Lett. 2, 869 (1990)

[7] N. Kagi, A. Oyobe, K. Nakamura, J. Lightwave Tech. 9, 261 (1991)

144

[8] J. W. Dawson, N. K. Park, K. J. Wahala, Appl. Phys. Lett. 60, 3090 (1992)

[9] N. K. Park, J. W. Dawson, K. J. Vahala, JEEE Photon. Tech. Lett. 4, 540 (1992)

[10] D. R. Wisely, Elec. Lett. 27, 521 (1991)

SSeS

aos